As of: 2026-11-21 | Global Security, Biotechnology and Artificial Intelligence | Strategic Horizon: 2027–2031

The convergence of artificial intelligence, biological research and organizational adaptation is challenging the historical assumptions underlying counterterrorism, biosecurity and national preparedness.

Scope: This assessment examines the lessons of Aum Shinrikyo’s biological weapons failures, the changing capabilities available to violent non-state actors, and the implications for global security, with particular attention to the United States, China, Japan, the European Union, Italy, France, Germany and the United Kingdom through 2031.

Executive Summary — Bottom Line Up Front

Principal judgment: Aum Shinrikyo’s unsuccessful biological weapons operations are no longer sufficient evidence that comparably motivated organizations will encounter the same barriers. Improvements in artificial intelligence may weaken some informational and organizational constraints that historically limited terrorist capabilities, although the extent to which these improvements translate into real-world biological weapons capability remains uncertain.

Aum Shinrikyo attempted biological attacks repeatedly before the 1995 Tokyo subway sarin attack. A historical study published by the US Centers for Disease Control and Prevention reported 10 unsuccessful attempts to conduct open-air biological attacks. The case demonstrates the importance of practical scientific competence, organizational discipline and reliable validation, rather than establishing that biological terrorism is intrinsically unachievable. CDC+1

The United Kingdom’s AI Security Institute has reported substantial advances in AI performance on specialized biological knowledge and laboratory-support evaluations. These findings establish improvements in measured capabilities, not equivalent improvements in a malicious actor’s probability of conducting a successful attack. AI Security Institute+1

The central emerging risk is the combination of technical assistance and organizational coordination: AI systems may help actors with some existing expertise access relevant information, distribute guidance, interpret failures and sustain complex projects. These mechanisms remain dependent on human judgment, physical resources, practical competence and information reliability.

Open-weight AI creates an additional governance problem because models can be redistributed, modified and operated outside the original developer’s access controls. The UK AI Security Institute identifies important limitations in existing safeguards. AISI Work+1

For governments, the strategic requirement is to move from predominantly actor-centered threat assessment toward integrated evaluation of intent, demonstrated capability, technology access, organizational behavior and opportunities for intervention.

The most consequential uncertainty is whether AI-assisted organizations can achieve sustained improvements in practical biological competence rather than simply producing more sophisticated plans or technical explanations.

The Next Bioterrorism Threat: Why AI Is Exposing the Limits of National Biosecurity

Artificial intelligence is advancing faster than governments can demonstrate the effectiveness of their biological safeguards. The consequences extend from scientific governance and pharmaceutical supply chains to intelligence coordination, industrial resilience and public finances.

The next biological security crisis may expose a fundamental contradiction in government preparedness: states are investing billions in surveillance, medical countermeasures and scientific infrastructure while the technologies influencing biological risk are changing faster than the institutions responsible for governing them. The United Kingdom’s 2026 Biological Security Strategy implementation report, the European Union’s Health Emergency Preparedness and Response Authority (HERA), and Washington’s May 2025 biological research executive order illustrate increasingly active but fragmented responses. Meanwhile, the UK AI Security Institute has documented advances in scientific capabilities that complicate traditional assumptions about access to specialist biological knowledge. The governing problem is no longer simply whether dangerous biological capabilities can be controlled, but whether public institutions can demonstrate that their safeguards, industrial resources and emergency systems remain effective as those capabilities evolve.

Aum Shinrikyo’s failure is becoming an unreliable foundation for security policy

Aum Shinrikyo remains one of the most revealing historical examples of the distance between terrorist ambition and operational capability. The Japanese movement possessed substantial resources, scientific personnel and an organizational structure capable of directing activity toward unconventional weapons. Yet a historical study published in Emerging Infectious Diseases in August 1999 documented at least ten unsuccessful biological attack attempts. The same organization demonstrated lethal chemical capabilities through the 1994 Matsumoto attack and the March 1995 Tokyo subway attack. The contradiction matters because it establishes that resources, technical qualifications and extremist determination do not translate uniformly into effective capabilities.

The deeper failure was organizational. Research published by the Center for a New American Security in December 2012 documented how Aum’s leadership structure, internal competition and uneven technical expertise complicated its weapons programmes. Scientific knowledge was not consistently matched to decision-making authority, while organizational incentives could reward the continuation of activities without reliable evidence of success. The historical lesson was therefore not that biological weapons were permanently beyond the reach of non-state actors, but that technical resources could be rendered ineffective by deficient institutional arrangements.

Other historical cases prevent that experience from becoming a universal rule. The 1984 Oregon salmonellosis outbreak associated with the Rajneeshee movement caused 751 identified illnesses, while the 2001 anthrax letter attacks in the United States resulted in 22 cases, including five deaths. The subsequent Amerithrax investigation required more than 10,000 witness interviews across six continents, according to the FBI’s historical case record. These incidents differed fundamentally in organization, consequences and investigative complexity. Together, they establish why biological terrorism cannot be evaluated through one presumed model of technical sophistication or one historical rate of success.

AI is changing scientific access, but governments still lack a reliable measure of operational consequences

The UK AI Security Institute’s Frontier AI Trends Report, published in 2025, identifies measurable progress in advanced models’ performance on specialized scientific evaluations. In its Biology QA and Chemistry QA assessments, human expert reference scores were 38% and 48%, respectively. Frontier models subsequently exceeded the relevant expert baselines. These percentages measure performance on defined evaluation tasks; they are neither estimates of biological weapons capability nor probabilities of successful terrorist attacks.

That distinction is economically and strategically consequential. A scientific benchmark measures what a model can demonstrate under specified conditions, while practical biological research depends on professional competence, reliable validation, physical resources and institutional oversight. Governments therefore face two opposing risks: treating advances in computational knowledge as evidence of operational capabilities that have not been demonstrated, or assuming that persistent laboratory constraints make those advances irrelevant to security policy.

The problem extends beyond individual expertise. AI systems can assist organizations in processing information, preserving institutional knowledge and supporting decisions, but their presence does not establish that those decisions become more reliable. The National Institute of Standards and Technology’s July 2024 Generative Artificial Intelligence Profile treats AI risk as a matter of governance throughout the system lifecycle, rather than simply model performance. In biological security, that distinction makes organizational accountability as important as technical evaluation: an institution using sophisticated AI can still produce poor judgments when its evidence is incomplete or its oversight ineffective.

Developers have begun responding. OpenAI published its biological capability preparedness approach in June 2025, while Anthropic activated AI Safety Level 3 protections in May 2025. Anthropic explicitly distinguished precautionary safeguards from confirmation that a capability threshold had been crossed. These initiatives demonstrate an institutional response to uncertainty, not proof that the relevant risks have materialized. The unresolved policy question is whether safeguards can be evaluated consistently across providers and remain effective as models and deployment arrangements change.

The industrial transmission mechanism runs through laboratories, supply chains and medical countermeasures

The biological security problem increasingly crosses the boundary between digital technology and physical industry. The World Health Organization’s September 2022 Global Guidance Framework for the Responsible Use of the Life Sciences assigns responsibility across governments, research institutions, scientists, funders, publishers and commercial organizations. That distribution reflects the structure of modern biotechnology: scientific activity may depend on several institutions, each responsible for a different component of research governance.

Artificial intelligence complicates this arrangement because centrally hosted models and open-weight systems operate under different forms of control. Providers of hosted services may be able to enforce access restrictions, update safeguards and intervene against identified misuse. Once open model parameters are independently distributed and modified, the original developer has considerably less control over downstream operation. The UK AI Security Institute has identified these limitations while recognizing the benefits of open models for scientific research and innovation. A restriction designed for one deployment model cannot automatically protect the wider biotechnology ecosystem.

The industrial consequences become tangible when prevention fails and governments must mobilize medical resources. HERA’s framework covers health-threat assessment, medical countermeasures, supply-chain monitoring and manufacturing preparedness. European Commission documentation identifies more than 500 prioritized medical countermeasure items across assessed threat categories, alongside approximately €1.7 billion of Horizon Europe funding earmarked for HERA within the 2021–2027 programme. The Commission also identified €39 million in direct grants supporting Member State sequencing capacity. These are different financial and operational categories: prioritized products are not necessarily procured stocks, earmarked funding is not equivalent to completed expenditure, and expanded diagnostic capacity does not establish comprehensive emergency readiness.

The distinction determines where industrial policy succeeds or fails. Procurement commitments may support manufacturing investment, but an announced requirement does not guarantee immediate production or delivery. Laboratories need qualified personnel; pharmaceutical manufacturers depend on production capacity and supply continuity; health systems require diagnostics, clinical resources and distribution arrangements. Biological preparedness therefore links public spending to the verified availability of physical capabilities. Without that connection, investment figures can overstate the resilience they are supposed to demonstrate.

Europe has national preparedness systems but no single operational authority

The European Union’s approach rests on coordination rather than the replacement of national health responsibilities. HERA contributes threat assessment and medical countermeasure preparedness, while Member States retain essential authority over healthcare organization and delivery. That division creates a permanent requirement for interoperability: European institutions may help identify needs and support common action, but national administrations must still turn those arrangements into medical services, procurement decisions and emergency operations.

Italy illustrates the consequences of territorial implementation. On 30 April 2026, the Ministry of Health announced approval of the National Pandemic Plan 2025–2029. On 21 May, the State–Regions Conference adopted the National Prevention Plan 2026–2031, accompanied by an additional €50 million for territorial prevention activities in 2026. These decisions establish national planning and financial commitments, but implementation depends on regions and autonomous provinces. The €50 million is not a dedicated counter-bioterrorism allocation, and the adoption of a national plan cannot establish equivalent preparedness throughout the Italian healthcare system.

France follows a different institutional model. The Secrétariat général de la défense et de la sécurité nationale coordinates interministerial nuclear, radiological, biological and chemical security arrangements, while the ORSAN framework organizes health-system responses to exceptional emergencies, including epidemic and biological risks. The advantage is an identifiable division between national security coordination and medical response. The constraint is operational: the existence of these structures does not demonstrate how effectively emergency services, scientific authorities and regional healthcare institutions would interact during a particular incident.

Germany combines specialized federal expertise with significant responsibilities at Länder and local levels. The Robert Koch Institute’s Centre for Biological Threats and Special Pathogens provides scientific capabilities relevant to biological incidents, including the assessment of exceptional pathogen-related threats. Its expertise strengthens national capacity, but the effectiveness of a distributed response depends on information exchange and cooperation among authorities exercising different responsibilities.

The United Kingdom offers a more explicit implementation record. Its June 2023 Biological Security Strategy established four functions: understand, prevent, detect and respond. The Cabinet Office’s July 2026 implementation report documented progress under that framework, while a parliamentary statement identified £1.83 billion of investment supporting National Biosecurity Centres and an additional £1 billion in health-protection measures associated with pandemic preparedness. The government also reported that its Biothreats Radar was being used by 14 departments and agencies. These amounts refer to separate commitments or programmes and should not be combined without reconciling their fiscal definitions. Nor does the reported deployment of a surveillance system independently establish its performance against every relevant biological threat.

The European comparison reveals a structural constraint rather than a clear national hierarchy. Italy relies heavily on regional execution, France on interministerial crisis coordination, Germany on federal scientific expertise operating within distributed governmental responsibilities, and the United Kingdom on a national strategy with published implementation reporting. None of these models can be declared superior from the available evidence. Their common vulnerability is the gap between formal preparedness and independently demonstrated operational performance.

Washington, Beijing and Tokyo are building different forms of biological security authority

Outside Europe, biological security is increasingly embedded in wider national security and scientific governance arrangements. The United States’ executive order of 5 May 2025, Improving the Safety and Security of Biological Research, directed changes to federal oversight of specified categories of research and called for stronger attention to biological security risks. It demonstrates executive policy direction, but the order itself does not prove completion of subsequent agency implementation. The distinction matters in a research economy spanning universities, federal laboratories, hospitals and private biotechnology companies, where federal funding rules and statutory oversight do not necessarily cover every activity in identical ways.

China’s Biosecurity Law, effective from 15 April 2021, provides a broader statutory framework covering infectious disease risks, biological research, laboratory safety, biological resources and threats involving biological terrorism and biological weapons. Its legal structure integrates biological security into national governance, but comprehensive legislative coverage is not a substitute for evidence of implementation. The relevant international comparison concerns the actual application of responsibilities, not the breadth of the statute’s language.

Japan has pursued institutional integration. On 1 April 2025, the Japan Institute for Health Security began operating through the combination of the National Institute of Infectious Diseases and the National Center for Global Health and Medicine. Its responsibilities bring together research, infectious disease intelligence, clinical expertise and emergency response. The institutional reform addresses coordination within Japan’s health security system, but its establishment does not by itself demonstrate complete operational maturity.

These national approaches operate within the framework established by the Biological Weapons Convention, whose prohibitions entered into force in 1975. The Convention remains the central international legal instrument against biological weapons, but it lacks a standing verification institution equivalent to the chemical weapons regime’s international organization. This limits the extent to which formal treaty participation can establish confidence in national implementation. International legal commitments, domestic oversight and operational preparedness therefore remain distinct requirements.

The next 24 months will expose the difference between expenditure and readiness

Between November 2026 and November 2028, the principal test will be whether governments can convert existing programmes into independently assessable capabilities. Italy’s National Prevention Plan 2026–2031 will require territorial implementation; the United Kingdom’s Biological Security Strategy will continue toward its 2030 objectives; HERA’s commitments under the 2021–2027 funding framework will move into a new programming period; and Japan’s Institute for Health Security will continue consolidating its integrated institutional functions. The relevant evidence will concern completed actions, documented exercises, available medical countermeasures and functioning information exchange, not additional announcements.

The fiscal constraint is already visible in the difference between the United Kingdom’s reported £1.83 billion investment in National Biosecurity Centres, its separate £1 billion health-protection measures, Italy’s additional €50 million for prevention in 2026 and HERA’s approximately €1.7 billion Horizon Europe allocation for 2021–2027. These figures cannot be treated as a common biological security budget: they cover different activities, jurisdictions and accounting periods. Their significance lies in what each programme delivers and whether public authorities can demonstrate that expenditure has improved preparedness.

For pharmaceutical manufacturers, biotechnology firms and scientific institutions, ineffective governance creates the risk of fragmented compliance requirements and poorly targeted restrictions. For health systems, inadequate coordination can translate into delayed access to diagnostics, medical resources and emergency support. For public finances, preparedness spending without verifiable outcomes creates exposure to repeated investment in capabilities whose practical performance remains uncertain. Governments also face a diplomatic cost when incompatible procedures obstruct cross-border cooperation under the Biological Weapons Convention and European health-security arrangements.

The cost of inaction will fall first on public institutions required to manage biological emergencies with incomplete information, then on taxpayers financing corrective measures and businesses exposed to disruption. The 2027–2031 outlook does not support a credible numerical forecast of AI-enabled bioterrorism, but it does establish a measurable test of government performance over the next 12–24 months: whether national authorities can demonstrate that legal controls, scientific evaluation, industrial preparedness and emergency response operate together. Without that evidence, higher spending and more elaborate strategies will remain commitments whose protective value has yet to be established.


Navigational Index — Three-Pillar Intelligence Dossier

PILLAR I — HISTORICAL FAILURE AND THE STRUCTURE OF BIOLOGICAL TERRORISM

Establishing the historical evidence, organizational constraints and distinctions between terrorist intention and operational capability.

  • Chapter 1. Aum Shinrikyo: Historical Record, Strategic Ambitions and the Limits of Biological Capability
  • Chapter 2. Expertise, Hierarchy, Secrecy and Organizational Failure in Violent Non-State Groups
  • Chapter 3. The Evolution of Biological Terrorism: Actors, Threat Patterns and the Limits of Historical Comparisons

PILLAR II — ARTIFICIAL INTELLIGENCE AND THE TRANSFORMATION OF BIOLOGICAL RISK

Assessing the measurable effects of technological change without conflating scientific progress with demonstrated weapons capability.

  • Chapter 4. Frontier AI, Biological Knowledge and the Changing Boundaries of Technical Expertise
  • Chapter 5. Organizational AI, Decision-Making, Knowledge Transfer and Adaptive Threat Networks
  • Chapter 6. The Dual-Use Biotechnology Ecosystem: Open Models, Commercial Services, Governance and Security Exposure

PILLAR III — GOVERNMENT RESPONSE, INTERNATIONAL SECURITY AND STRATEGIC OUTLOOK

Connecting the evolving threat to international law, national security institutions, economic resilience and defensible policy choices.

  • Chapter 7. National and Multilateral Biosecurity: The United States, China, Japan, Italy, France, Germany, United Kingdom and European Union
  • Chapter 8. Strategic Scenarios for 2027–2031, Early-Warning Indicators, Policy Options and Final Net Assessment

Master Abstract

Historical failure is evidence of constraints, not proof of permanent immunity

The central analytical problem is the distinction between a historical observation and the general conclusion subsequently drawn from it. Aum Shinrikyo possessed substantial organizational resources, ideological motivation and access to scientifically educated personnel, yet its biological attacks failed to produce the intended effects. In contrast, the organization demonstrated a capacity for lethal chemical violence in Japan. The difference between these outcomes shows that possession of resources and determination does not automatically confer competence across different technical domains. It also demonstrates why analysts should distinguish an organization’s declared ambitions from its demonstrated ability to execute them.

The 1999 CDC-hosted historical review documented Aum Shinrikyo’s unsuccessful biological attacks and emphasized technical barriers facing terrorist organizations. Its analysis also recognized that the diffusion of dual-use technologies could erode some of those barriers. The historical evidence therefore supports two propositions simultaneously: biological weapons development remains technically demanding, and those demands cannot be assumed to remain constant over time.

This distinction matters because governments evaluating a potential biological threat must determine whether a group can transform motivation, personnel and resources into an effective capability. A historical failure may reveal constraints affecting one organization at one point in time. It does not necessarily establish the permanence of those constraints across different organizations, technological environments or institutional arrangements.

Artificial intelligence changes the analytical baseline, but its operational effects remain incompletely measured

The development of general-purpose and specialized biological AI systems introduces a new variable into assessments of non-state weapons proliferation. Advanced models can increasingly perform tasks involving scientific reasoning, information synthesis and laboratory-support knowledge. The United Kingdom’s AI Security Institute, drawing on frontier-model evaluations undertaken since November 2023, reports that advanced systems now surpass expert human baselines on certain biology and chemistry assessments and demonstrate increasingly capable experimental troubleshooting.

These results are significant because they challenge the assumption that specialist knowledge can only be made available through the recruitment and retention of suitably trained personnel. However, the distinction between model performance and operational capability remains fundamental. High scores on scientific evaluations do not establish that a model can independently conduct reliable laboratory work, provide correct guidance under uncontrolled conditions or enable the production of an effective biological weapon. Even accurate information must be interpreted and applied within the constraints imposed by physical systems, materials, human expertise and organizational decision-making.

The most important question is therefore not whether AI eliminates traditional barriers to biological terrorism, but which barriers it might weaken, how far, for whom, and under what conditions. This calls for empirical evaluation of AI-assisted performance compared with realistic alternative sources of information, rather than assessments based exclusively on benchmark scores or speculative capability projections.

Organizational adaptation may be as consequential as access to scientific information

The potential effect of AI extends beyond the individual user’s ability to obtain an answer. Organizations can employ digital systems to manage institutional knowledge, standardize reporting, support coordination and interpret incomplete information. These functions are widely relevant in legitimate enterprises and scientific institutions, but their possible adoption by violent non-state actors creates an additional security consideration.

In this context, AI could change the relationship between leadership authority and technical expertise. An organization that historically depended on a narrow group of specialists may become less dependent on individual knowledge holders for certain information-processing functions. Nevertheless, such systems also introduce new weaknesses: unreliable recommendations, automation bias, manipulated information, false confidence and failures of human oversight. An organization that mistakes apparent technical sophistication for reliable validation may reproduce historical errors through a different technological mechanism.

The principal analytical hypothesis is that AI-enabled organizational adaptation could be consequential even where the technology itself does not directly overcome the most difficult biological barriers. This hypothesis remains conditional. Demonstrating it requires credible evidence of actual improvements in organizational performance and relevant capabilities, not merely accounts of AI adoption.

Biological security is becoming a problem of cross-sector governance

The governance challenge is amplified by the different arrangements through which AI systems are distributed. Providers of centrally hosted models can apply access restrictions, monitor misuse within applicable legal and privacy constraints, and update safeguards. Open-weight systems present different trade-offs because their parameters can be redistributed and modified after release. The UK AI Security Institute has specifically identified the difficulty of maintaining robust safeguards for such systems, while recognizing their benefits for scientific research, transparency and innovation.

Biological security consequently cannot be reduced to AI model moderation. It also depends on responsible research governance, physical security, lawful controls over sensitive materials, supply-chain oversight, public-health surveillance, microbial forensics and emergency preparedness. The World Health Organization emphasizes that deliberate biological events create both public-health and security responsibilities, requiring coordination among medical, epidemiological, law-enforcement and investigative authorities. Organizzazione Mondiale della Sanità

For Europe, this challenge intersects with the European Commission’s Preparedness Union Strategy and the work of its Health Emergency Preparedness and Response Authority. For the United Kingdom, the Biological Security Strategy offers an explicit framework spanning understanding, prevention, detection and response. These are relevant institutional foundations, but their effectiveness against an evolving AI-enabled threat requires examination of implementation, interoperability and measurable readiness rather than policy announcements alone.

Key Evidence Table — Verified Baseline

IndicatorValue or established statusReference dateDefinition and scopeIssuer and source
Aum Shinrikyo biological operations10 unsuccessful open-air attack attempts reportedHistorical events before 1995; analysis 1999Historical case-study finding, not a current global incident rateHistorical Trends Related to Bioterrorism — CDC-hosted journal, 1999
Frontier AI evaluationsEvaluation programme operating since November 2023Report published December 2025Assessed frontier AI capabilities across multiple security-relevant domainsFrontier AI Trends Report — UK AI Security Institute, 2025
Biology-related AI performanceAbove expert baselines on selected evaluations2025 reportBenchmark performance, not proof of successful malicious biological operationsFrontier AI Trends Report — UK AI Security Institute, 2025
Open-weight model governanceSafeguard robustness and post-release control remain unresolved challenges2026Model distribution, modification and misuse-risk managementManaging Risks from Increasingly Capable Open-Weight AI Systems — UK AISI, 2026
UK national biological securityFour strategic functions: understand, prevent, detect and respondJune 2023National policy architecture and 2030 resilience objectivesUK Biological Security Strategy — Cabinet Office, June 2023
EU health-threat preparednessPandemic threats, CBRN threats and antimicrobial resistance identified as priority areasPrioritization initiated 2022; continuing frameworkMedical countermeasure planning and health-threat intelligenceThreat Assessment and Intelligence Gathering — European Commission HERA
International biological incident coordinationJoint health and security response requiredApril 2024Preparedness and response to deliberate biological and other hazardous eventsDeliberate Events — World Health Organization, April 2024

Interpretation: These records establish the historical baseline, improvements in measured AI capabilities and the existence of relevant security institutions. They do not establish a numerical increase in the probability of AI-enabled bioterrorism or demonstrate that contemporary extremist organizations possess effective biological weapons capabilities.

Competing Explanations and Future Pathways

Three principal pathways are analytically plausible. They can overlap and should not be treated as mutually exclusive outcomes.

PathwayDiagnostic supportCountervailing evidence or limitationObservable indicatorsCurrent assessment
A. Persistent physical and organizational constraintsHistorical failures and the continued need for practical laboratory skills, reliable processes and physical resourcesAI and biotechnology may reduce some information-related constraintsIndependent evidence of persistent real-world failure despite increased access to AIA credible baseline; the magnitude of change remains unresolved
B. Selective AI-enabled capability improvementAdvances in AI scientific reasoning and laboratory-support evaluationsBenchmark results do not establish reliable real-world upliftReproducible, safely conducted comparative studies showing improved outcomes for legitimate laboratory usersStrong evidence of improving measured capabilities; limited evidence of the resulting security impact
C. AI-enabled organizational adaptationAI can support knowledge management, decision assistance and coordination in general organizational settingsSpecific effects on violent networks and biological weapons capability are insufficiently establishedCorroborated investigations of AI’s organizational use, supported by independently validated performance evidenceConsequential hypothesis requiring additional empirical validation

The second and third pathways warrant sustained evaluation because their implications are not captured by traditional assessments focused solely on the number of formally trained specialists within an organization. Yet the first pathway remains an essential counterweight: technical information is not a substitute for practical competence, and AI systems can compound mistakes when their outputs are accepted uncritically.

It would therefore be premature either to conclude that advanced AI has transformed bioterrorism into an easily accessible capability or to assume that historical failure rates will remain stable irrespective of technological change.

Principal Intelligence Gaps and Watch Indicators

The gap between AI evaluation and real-world capability

The most important evidentiary gap concerns practical uplift. Government evaluations demonstrate progress in selected scientific tasks, but those findings cannot be converted directly into estimates of terrorist operational success. Decision-relevant research should distinguish improved scientific comprehension, improved performance by already qualified specialists, and demonstrated completion of complex tasks in controlled, legitimate laboratory settings.

A useful monitoring system would track the methodological quality, reproducibility and external validity of such evaluations, including whether performance improvements persist when AI-generated recommendations encounter real-world uncertainty.

The gap between organizational adoption and operational effectiveness

The presence of AI tools within an organization is not proof that those tools are useful. Investigations need to distinguish reported adoption, frequency of use, actual changes in decision-making, and independently demonstrated operational consequences. Where security agencies identify extremist use of AI, information should be assessed within lawful investigative procedures and not converted into unsupported assumptions about biological capabilities.

The limits of safeguard effectiveness

Model providers have different opportunities for intervention depending on their deployment and distribution arrangements. Hosted systems may permit monitoring and enforcement, whereas independently operated open-weight systems are harder for their original developers to oversee.

The decision-relevant indicators are not only whether a model refuses a prohibited request in a laboratory evaluation, but whether safeguards remain effective under authorized adversarial assessment, whether discovered weaknesses are remedied, and whether relevant governance arrangements can adapt to new capabilities.

The institutional coordination gap

Deliberate biological threats can first appear as public-health anomalies rather than immediately recognizable security incidents. This creates challenges for attribution, information exchange, clinical response and coordination between civilian and security institutions.

The WHO’s approach to deliberate events establishes the need for integrated health-security arrangements. The European Commission’s preparedness framework likewise emphasizes cross-government coordination, crisis planning and strengthened response capacity. Organizzazione Mondiale della Sanità+1

For the European country assessments, the consequential question will be whether Italy, France, Germany and the United Kingdom have sufficiently compatible procedures to exchange relevant warnings, characterize suspected events, coordinate public communication and mobilize countermeasures without undermining scientific research or civil liberties.

The absence of a defensible numerical forecast

Available public evidence does not support a reliable numerical probability that an AI-enabled organization will conduct a successful mass-casualty biological attack within the 2027–2031 horizon.

Any quantified forecast would require a defined outcome, usable incident base rates, consistent treatment of unsuccessful attempts and credible measures of the relationship between AI assistance and practical biological capability. The current evidence establishes reasons to re-examine the threat model, not a scientifically justified numerical forecast of attack frequency.

Strategic Decision Implications

The historical record offers an important lesson for contemporary counterterrorism. Aum Shinrikyo demonstrated that substantial resources, motivation and scientific personnel do not guarantee the successful development of biological weapons. Modern AI evaluations demonstrate that access to some forms of scientific knowledge and support is changing. Neither finding, considered independently, establishes the future trajectory of biological terrorism.

The policy challenge lies in integrating these observations without exaggerating or dismissing the threat.

Governments need assessments that distinguish intent from capability, information access from practical competence, and AI adoption from proven operational effectiveness. They also need institutions capable of detecting biological emergencies, coordinating health and security responses, maintaining proportionate safeguards and preserving the benefits of legitimate biotechnology research.

The most credible strategic response is therefore layered: evidence-based model evaluations, responsible biological research governance, effective public-health surveillance, secure handling of sensitive materials, lawful intelligence collection and regularly exercised emergency preparedness. These measures should be judged against measurable outcomes, implementation costs, scientific consequences and civil-liberties considerations.

The decisive issue is not whether future terrorists will possess more advanced tools than Aum Shinrikyo. It is whether technological and organizational changes will materially narrow the distance between malicious intention and demonstrated capability—and whether national and international security institutions can identify that change before it produces irreversible consequences.


BIOSECURITY INTELLIGENCE
STRATEGIC THREAT ASSESSMENT · 21 NOVEMBER 2026

BIOTERRORISM IN THE AGE OF ARTIFICIAL INTELLIGENCE

Historical failure, emerging technological capabilities and the transformation of biological security governance.

GLOBAL SECURITY 2027–2031 HORIZON QUALITATIVE ASSESSMENT OPEN-SOURCE EVIDENCE

Strategic Dependency Network

Select a node to examine its security implications.

INTERACTIVE MODEL
Historical failure

Aum Shinrikyo’s unsuccessful biological operations demonstrate serious technical and organizational constraints. They do not prove these barriers are permanent.

Three-Pillar Strategic Architecture

Explore the evidence, technological transition and government response.

Historical Failure

The historical record establishes that resources and motivation do not automatically translate into effective biological weapons capability. Practical expertise, organizational decision-making and validation matter.

KEY INTELLIGENCE JUDGMENT

Past operational failure should inform present risk assessment, not determine it.

Evidence and Uncertainty Matrix

Qualitative classifications; no fabricated risk scores.

DOMAIN ASSESSMENT EVIDENCE STATUS POLICY IMPLICATION
Historical biological attacks Repeated Aum Shinrikyo failures Documented historical finding Maintain technical and organizational distinctions
AI scientific capabilities Performance improvements in selected evaluations Measured on evaluated tasks Continue controlled capability assessment
AI-enabled operational uplift Magnitude remains uncertain Insufficient real-world evidence Avoid unsupported probability estimates
Open-weight safeguards Limited post-release developer control Documented governance challenge Evaluate safeguards before and after release
Emergency preparedness Integrated public-health and security response Established institutional requirement Test coordination and response readiness

Historical and Strategic Timeline

Documented events and future assessment horizon are explicitly distinguished.

1990s
Aum Shinrikyo’s biological operations fail to achieve their intended effects.
2023
The United Kingdom publishes its national Biological Security Strategy.
2025
UK AI Security Institute reporting documents advances on selected frontier AI evaluations.
2026
The assessment examines open-weight governance, safeguards and emerging capability uncertainty.
2027–31
Prospective monitoring horizon, not a prediction of an attack or its probability.

Strategic Net Assessment

From historical constraints to adaptive security

Artificial intelligence may change access to specialist information and how organizations coordinate expertise. These changes do not establish effective biological weapons capabilities. Governments should evaluate demonstrated performance, maintain layered safeguards and strengthen integrated public-health and security preparedness.

OPEN-SOURCE ANALYTICAL ASSESSMENT | 21 NOVEMBER 2026

PILLAR I — HISTORICAL FAILURE AND THE STRUCTURE OF BIOLOGICAL TERRORISM

Historical evidence, organizational constraints, terrorist capability formation and the limits of retrospective threat assessment

Chapter 1. Aum Shinrikyo: Historical Record, Strategic Ambitions and the Limits of Biological Capability

The Japanese precedent and the distinction between ambition and capability

Aum Shinrikyo occupies an unusual position in the history of terrorism because its activities exposed the gap between an organization’s capacity to mobilize substantial financial, human and technical resources and its ability to translate those resources into reliable unconventional weapons capabilities. The movement, founded by Shoko Asahara in Japan and transformed from a religious organization into an increasingly authoritarian and violent structure, developed ambitions extending far beyond its original spiritual identity. Its leadership embraced an apocalyptic worldview, systematically targeted perceived enemies and eventually undertook activities involving chemical and biological weapons. Its importance to contemporary security analysis lies not merely in the scale of those ambitions but in the extraordinarily uneven outcomes of its different programmes, which make it possible to examine the relationship between organizational authority, scientific expertise, operational performance and institutional failure within the same violent non-state organization.

The historical record establishes that Aum Shinrikyo repeatedly attempted biological attacks without achieving its intended effects. Jonathan Tucker’s study, published in the CDC’s Emerging Infectious Diseases journal in August 1999, documented at least ten attempts to use biological agents. The same organization nevertheless demonstrated lethal capability through chemical attacks, illustrating that technical performance in one field cannot be treated as evidence of equivalent competence in another. The historical investigation distinguished confirmed incidents from allegations, plans and hoaxes, an important methodological discipline because the available record of biological terrorism contains substantial inconsistencies in both reporting and classification. Historical Trends Related to Bioterrorism: An Empirical Analysis — Emerging Infectious Diseases, August 1999. CDC

The implications are more significant than the familiar observation that biological weapons are difficult to develop. Aum demonstrates why the possession of scientists, laboratories, financing and political or religious commitment cannot be added together to produce a meaningful measure of operational capability. Those resources constitute potential inputs to a programme; they do not establish that its leadership can select qualified personnel, make technically sound decisions, recognize failure, maintain reliable processes or determine whether an undertaking has achieved its intended objective. The distinction is central to intelligence assessment because an organization may possess the appearance of sophistication without the organizational arrangements required to use that sophistication effectively. Conversely, an organization with fewer visible resources may sometimes exploit superior discipline, specialized knowledge or external assistance more efficiently. Neither proposition supports assuming that any particular contemporary group possesses effective biological weapons capabilities.

The transition from religious authority to organized political violence

Aum Shinrikyo’s transformation illustrates how ideological authority can reshape the decision-making structures of an organization until loyalty to leadership becomes inseparable from its operational objectives. Asahara’s position was not equivalent to that of an ordinary organizational executive whose decisions could be challenged through established technical or administrative procedures. The movement’s internal structure elevated his authority while reducing the independence of subordinate judgments. This distinction matters because unconventional weapons development requires repeated assessments of evidence, and those assessments may contradict the expectations, preferred schedules or ideological commitments of senior leaders.

The Japanese Public Security Intelligence Agency’s historical account identifies the 1994 Matsumoto sarin attack and the 20 March 1995 Tokyo subway attack among Aum Shinrikyo’s major acts of violence. Its retrospective publication reports eight deaths and 140 injuries in Matsumoto, based on criminal trial records, and 13 deaths and more than 5,800 injured people in Tokyo, using the figures established for the relevant Japanese victim-relief arrangements. These figures differ from some contemporary reports and earlier historical studies, which used different reporting dates and casualty definitions. Such differences should not be silently reconciled into a single supposedly universal figure. 25 Years After the Tokyo Subway Sarin Gas Attacks — Public Security Intelligence Agency, Japan. 公安調査庁

The discrepancy between the group’s unsuccessful biological activities and its destructive chemical attacks establishes a fundamental point about terrorist capability formation: it is possible for an organization to possess both genuine operational competence and profound technical deficiencies at the same time. A threat assessment that classifies Aum simply as technically sophisticated would obscure the failure of its biological programme, while an assessment describing it as technically incompetent would ignore its demonstrated ability to commit lethal chemical violence. The analytical unit must therefore be the specific capability, supported by evidence of performance, rather than the organization’s overall reputation or the qualifications of its most prominent members.

Interpreting the biological programme without overstating the evidence

Historical accounts of Aum’s biological activities describe substantial investments of attention and resources, but the outcomes do not support treating these efforts as a proven biological weapons capability. A group may investigate a hazardous agent, acquire scientific equipment or undertake experimental work without establishing that it has developed a reliable means of causing the intended harm. Similarly, a reported attempt does not become evidence of technical success merely because the organization believed it was conducting an operation. The correct distinction is between evidence of intention, evidence of preparation, evidence of attempted action and evidence of actual effects.

The study published in Emerging Infectious Diseases provides a useful example of why this distinction matters. Its historical database included multiple kinds of incidents, ranging from threats and acquisition attempts to confirmed use, and the author explicitly warned against unreliable or sensationalized accounts. The study’s treatment of Aum is accordingly valuable not because it resolves every technical question concerning the movement’s activities, but because it places the group’s record within a broader attempt to distinguish documented events from allegations that had entered public discussion without sufficient confirmation. Historical Trends Related to Bioterrorism — CDC-hosted journal, August 1999, Table 1. CDC

The subsequent investigation by Richard Danzig and colleagues, published through the Center for a New American Security in December 2012, adds a different evidentiary perspective through retrospective research involving former members and detailed examination of the organization’s internal activities. This material is particularly useful for understanding reported decision-making and internal organizational dynamics, although retrospective testimony must be evaluated with attention to imperfect recollection, individual interests and the difficulty of independently corroborating statements made years after the events. It should not be treated as equivalent to a contemporary physical investigation or a judicial finding merely because it supplies a detailed account. Aum Shinrikyo: Insights Into How Terrorists Develop Biological and Chemical Weapons, Second Edition — Center for a New American Security, December 2012. CNAS

The limits of technical specialization

A persistent analytical weakness in discussions of unconventional weapons is the assumption that scientific education establishes broadly transferable operational expertise. Scientific disciplines are highly specialized, and competence in one field does not automatically provide the practical or managerial knowledge necessary for a complex programme extending across several fields. Furthermore, laboratory knowledge, engineering competence, quality assurance and the ability to interpret uncertain experimental results are distinct functions. Organizations that treat them as interchangeable may obtain little practical benefit from personnel who are otherwise highly educated.

This observation does not mean that scientific qualifications are irrelevant. It means that the correct assessment concerns the relationship between individual expertise and the actual responsibilities assigned to personnel. The important questions are whether the organization recognizes differences between specialists, whether technically competent individuals can influence decisions, whether leadership distinguishes uncertain findings from verified results, and whether organizational procedures allow erroneous assumptions to be challenged. These questions can be investigated without reconstructing sensitive methods of biological weapons development or attempting to identify specific technical vulnerabilities that could be exploited for malicious purposes.

Aum’s biological failures therefore provide evidence about the limits of a particular organizational and technical arrangement, rather than evidence of an immutable scientific barrier. The stronger historical conclusion is that scientific resources become operationally meaningful only when they are embedded within an organization capable of using them competently. This conclusion applies to legitimate scientific institutions as well, although the security consequences differ fundamentally according to the purpose for which capabilities are being developed.

The intelligence significance of contrasting outcomes

The contrast between Aum’s chemical and biological activities also cautions against interpreting failure as reassurance. An unsuccessful programme may indicate technical incapacity, defective organizational arrangements, external disruption or a combination of those factors. Historical evidence cannot always establish the contribution of each factor with sufficient precision to support a universal explanation. The same organization can exhibit different strengths and weaknesses in different activities, and its future behavior may change following personnel replacement, leadership decisions or technological developments.

Equally, the existence of several unsuccessful attempts should not automatically imply that success was becoming more likely. Repetition is not itself evidence of learning. Without reliable evidence of corrected deficiencies, improved judgment and better organizational processes, successive activities may merely reproduce earlier mistakes. The distinction between repetition and demonstrable improvement is therefore critical when assessing non-state actors that announce technical ambitions, claim new capabilities or display unfamiliar equipment.

The institutional implication is that security agencies should distinguish an organization’s declared intentions, accessible resources, demonstrated performance, unresolved deficiencies and ability to change. Aggregating these categories into a single threat label risks confusing the seriousness of an actor’s intentions with the maturity of its capabilities. Aum remains a particularly valuable historical case precisely because it exposes this error.

Chapter 1 — Key judgments

Aum Shinrikyo’s record demonstrates the coexistence of extreme violent intent, substantial organizational resources, unsuccessful biological operations and demonstrated lethal chemical violence. The evidence supports a capability-specific approach to intelligence assessment, in which technical performance is judged according to the relevant activity rather than the organization’s general reputation. Historical evidence of biological failure remains important, but it cannot establish that future organizations will experience identical constraints under different technological and organizational conditions.

What would change this assessment: Newly authenticated contemporary operational records, independently corroborated testimony or forensic findings that materially alter the established chronology, outcomes or organizational interpretation of Aum’s activities.

Open official record: The most consequential limits concern the completeness of historical internal records and the extent to which retrospective witness accounts can be corroborated against contemporaneous Japanese investigative and judicial documentation.

Chapter 2. Expertise, Hierarchy, Secrecy and Organizational Failure in Violent Non-State Groups

Why technical resources can fail to produce organizational competence

The analytical significance of organizational structure becomes apparent when technical resources exist but the institution cannot use them effectively. A hierarchical organization may possess qualified individuals while systematically preventing them from exercising independent judgment, communicating inconvenient findings or recommending the suspension of an unsuccessful undertaking. In such circumstances, the institution’s apparent concentration of authority becomes a source of fragility rather than effectiveness. Decisions can be taken rapidly, but speed is not equivalent to accuracy, and centralized control may reduce the organization’s capacity to detect its own mistakes.

The Aum Shinrikyo case demonstrates why the relationship between knowledge and authority deserves independent consideration. Its internal structure combined ideological obedience with an operational system in which seniority and proximity to leadership did not necessarily correspond to technical expertise. Danzig and his collaborators documented the movement’s difficulties in translating available resources into reliable outcomes and examined how individual responsibilities, leadership expectations and organizational pressures shaped those results. Their findings provide evidence for examining how authority can distort technical judgment, while leaving room for uncertainty about the precise contribution of each reported organizational weakness. Aum Shinrikyo: Insights Into How Terrorists Develop Biological and Chemical Weapons — CNAS, December 2012. CNAS

For institutional analysis, four distinct failures must be separated. The first concerns access to relevant expertise, the second concerns the authority assigned to technically competent personnel, the third concerns the reliability of information communicated inside an organization, and the fourth concerns whether leaders respond appropriately when evidence contradicts their assumptions. These are connected problems, but they are not identical. An organization might recruit qualified specialists yet still make poor decisions; it might allow independent judgment but restrict the exchange of relevant information; or it might receive accurate information while maintaining incentives that discourage corrective action.

The conflict between centralized authority and independent verification

Centralized leadership offers obvious administrative advantages. Senior authorities can establish priorities, allocate resources and coordinate departments without prolonged negotiation. These characteristics, however, become counterproductive when leaders treat their own expectations as superior to observations made by qualified personnel. The problem is especially pronounced in organizations where authority is legitimized through personal loyalty, ideological doctrine or privileged access to a dominant leader rather than accountable institutional procedures.

The central issue is not whether leadership should delegate every decision. It is whether authoritative decisions remain subject to meaningful verification. Complex technical undertakings require the ability to distinguish an established result from an assumption, recognize disagreement among qualified individuals and revise a course of action when evidence changes. Where such procedures are absent, an organization may confuse the existence of activity with the achievement of progress. Financial expenditure, personnel mobilization and visible technical infrastructure can then become symbols of success even when there is insufficient evidence that the underlying capability functions as intended.

From an intelligence perspective, this distinction argues against interpreting organizational discipline solely through observable obedience or the existence of a formal command structure. An organization can be highly disciplined in enforcing leadership directives while remaining unreliable in its assessment of technical information. Conversely, visible internal disagreement does not necessarily indicate weakness if the institution possesses legitimate processes through which competing findings can be evaluated.

Secrecy, compartmentalization and the loss of institutional knowledge

Secrecy is an inherent feature of clandestine organizations, but it creates costs that are frequently obscured by the assumption that restricted information automatically improves security. Compartmentalization can reduce the number of people exposed to sensitive activities, yet the same restrictions may prevent relevant expertise from being brought into discussions, fragment institutional awareness and limit opportunities to correct mistakes. The problem becomes particularly severe when individuals responsible for decisions are unable to consult those who possess the knowledge necessary to evaluate their consequences.

The analytical lesson extends beyond the number of people who hold information. Organizations require some mechanism for determining whether information is accurate, whether uncertainties have been communicated, and whether conflicting assessments can reach decision-makers. A compartmentalized organization may prevent disclosure while simultaneously depriving itself of the means to evaluate its own activities. These costs can remain hidden because external observers see neither the unavailable expertise nor the decisions that could have changed had that expertise been consulted.

The effect on threat assessment is considerable. Intelligence agencies may identify scientific personnel, facilities or organizational divisions without knowing whether meaningful communication exists among them. It would be inappropriate to assume that the presence of those resources establishes a coherent programme. Equally, limited visibility into an organization’s internal communication is not sufficient evidence that its coordination is defective. A distinction must therefore be maintained between observed organizational characteristics and inferred consequences that cannot yet be demonstrated.

Internal competition and the distortion of reported performance

Organizations organized around personal authority may create incentives for subordinate leaders to associate themselves with apparently successful projects, compete for privileged attention and avoid reporting negative results. Such incentives do not require individuals to be scientifically incompetent or deliberately deceptive in every instance. They can emerge when promotion, access to resources or continued organizational influence depend on demonstrating loyalty and producing favorable reports. Under these conditions, information transmitted to leadership may become increasingly selective, especially when reporting disappointing outcomes carries personal consequences.

In its study of Aum Shinrikyo, the CNAS research team explored the relationship between internal hierarchy, competition and the organization’s technical undertakings. The broader institutional significance is that a weapons programme may develop a political function within its parent organization, independent of whether it is achieving the stated technical objective. A programme can continue to attract support because it symbolizes leadership commitment, confers institutional status or reinforces expectations that the organization is making progress. These pressures may make it difficult for decision-makers to distinguish genuine advancement from activity maintained for organizational reasons. Aum Shinrikyo — CNAS, December 2012. CNAS

The implication for external assessment is that claims made by violent organizations must be evaluated according to their evidentiary value rather than their rhetorical confidence. Statements intended to reassure followers, impress rivals or influence public opinion may reveal political objectives while providing little reliable evidence about technical maturity. An organization can exaggerate its capabilities, misunderstand them or conceal difficulties. In each case, the resulting public claim should be distinguished from independently established performance.

Learning, repetition and the interpretation of failure

Institutional learning requires more than the ability to repeat an activity after an unsuccessful result. An organization must possess sufficiently reliable information about what happened, recognize the limitations of that information and incorporate valid findings into future decisions. Without these conditions, repetition may become a substitute for analysis. The mere occurrence of subsequent attempts does not establish that earlier weaknesses were understood or corrected.

This distinction is particularly important when comparing historical terrorism cases. Some organizations may change their activities after failure, but a change in visible behavior does not by itself demonstrate that the institution acquired more reliable technical knowledge. Changes can result from leadership turnover, resource depletion, changes in political opportunity, external pressure or a reassessment of priorities. Assessing whether organizational learning occurred requires evidence that supports a connection between earlier experience and subsequent improved performance.

For government analysts, the practical question is whether a reported organizational change corresponds to a demonstrable difference in capability or merely to a new claim, structure or activity. The evidentiary threshold should remain demanding because both underestimation and overestimation carry costs. Underestimation can obscure a genuine emerging threat; overestimation can encourage disproportionate security measures, misdirect public resources and create incentives for violent actors to exploit the publicity generated by their claims.

The organizational dimension of artificial intelligence

Although the detailed technological assessment belongs to Pillar II, the historical discussion establishes why artificial intelligence should eventually be evaluated as more than a source of technical information. Organizations use information systems to structure internal communication, preserve knowledge, support administrative decisions and examine performance. The potential security implications arise when technological systems change how an organization distributes authority, interprets evidence or manages institutional memory.

However, technological adoption does not automatically resolve the problems identified in the Aum case. An AI system can deliver inaccurate information, present uncertainty inadequately or encourage excessive reliance on seemingly authoritative answers. If leadership substitutes confidence in algorithmic output for genuine independent verification, the organization may preserve its original weaknesses in a different form. The relevant distinction is therefore between technological availability and demonstrable institutional improvement. The existence of more advanced information-processing tools cannot establish that an organization will use them reliably.

This issue is important for institutional warning systems because intelligence collection focused only on scientific credentials, finances or physical facilities may fail to capture changes in how an organization manages knowledge. Yet the opposite error must also be avoided: observing AI adoption should not be interpreted as proof that a violent organization has acquired advanced biological capabilities. Evidence of software use and evidence of practical weapons capability belong to different categories.

Chapter 2 — Key judgments

Organizational authority, internal incentives and information quality can influence whether technical resources are used effectively. Historical evidence from Aum Shinrikyo supports examining the relationship between scientific expertise and decision-making power, rather than treating personnel qualifications or equipment ownership as sufficient evidence of capability. The principal contemporary implication is that threat assessment should distinguish organizational adaptation from independently demonstrated technical improvement.

What would change this assessment: Reliable organizational records showing that particular decision-making arrangements systematically improved or degraded technical performance, supported by corroborated evidence rather than claims of expertise or organizational sophistication.

Open official record: Detailed, independently verifiable documentation of internal organizational processes remains limited for many clandestine groups. The absence of such evidence should be treated as an intelligence gap, not as proof of competence or dysfunction.

Chapter 3. The Evolution of Biological Terrorism: Actors, Threat Patterns and the Limits of Historical Comparisons

Why the historical record cannot be interpreted as a single trajectory

The development of biological terrorism cannot be adequately understood through one organization, one ideological movement or one category of biological incident. Historical cases differ substantially in motivation, organizational structure, availability of scientific expertise, intended effects, actual consequences and the quality of the evidence available to investigators. Consequently, an account that moves directly from Aum Shinrikyo’s failures to a general prediction about future biological terrorism would overlook some of the most consequential differences within the historical record. The central task is to establish which observations can legitimately be compared, which belong to fundamentally different categories and which remain too uncertain to support conclusions about broader patterns of capability.

The study published by Jonathan Tucker in August 1999 provides an important historical baseline, although it should not be treated as a contemporary threat database. The underlying Monterey database contained 415 reported incidents involving chemical, biological, radiological or nuclear materials through 31 January 1999. Of these, 151 were categorized as terrorist incidents with sufficient information to permit comparison, while 33 involved biological agents within the relevant terrorist subset. These figures refer to different nested categories and cannot be presented as interchangeable measures of biological attacks. The original study also acknowledged that the database depended on reported incidents, creating substantial limitations concerning undetected activity, incomplete records, inconsistent reporting and the changing attention of law-enforcement agencies and the media. Historical Trends Related to Bioterrorism: An Empirical Analysis — Emerging Infectious Diseases, August 1999. CDC

The distinction between the 415 recorded incidents and the 33 biological cases is especially important because apparently precise historical numbers can create misleading impressions of statistical completeness. An incident involving a threat does not establish that an actor possessed dangerous biological material, and possession does not establish that an effective biological weapon existed. A hoax can generate considerable public concern without involving a biological agent at all. An attempted attack may reveal willingness to act without demonstrating the ability to cause harm. A successful biological incident, meanwhile, may involve localized consequences and a relatively small organizational footprint rather than the mass-casualty scenario that dominates public discussion of unconventional terrorism. Treating all these categories as observations of the same underlying phenomenon would obscure the information necessary for reliable policy analysis.

The 1984 Oregon incident and the reality of localized biological violence

The 1984 outbreak in The Dalles, Oregon, associated with members of the Rajneeshee movement, provides an important counterexample to the proposition that violent non-state actors have never successfully caused disease through deliberate biological contamination. A retrospective epidemiological investigation, published in the Journal of the American Medical Association in August 1997, identified 751 cases of salmonellosis associated with a community outbreak between September and October 1984. The investigation used clinical and epidemiological evidence to establish the outbreak’s scale and relationships between cases, rather than relying exclusively on later accounts of the perpetrators’ intentions. A Large Community Outbreak of Salmonellosis Caused by Intentional Contamination of Restaurant Salad Bars — JAMA, August 1997, indexed by the National Library of Medicine. PubMed

The incident demonstrates that biological violence can occur without resembling the technologically ambitious programmes associated with mass-casualty unconventional weapons. Its political setting, organizational motivations and consequences differed substantially from those of Aum Shinrikyo. It also shows why the distinction between the occurrence of a deliberate biological incident and the existence of a sophisticated biological weapons programme must be preserved. The former may be established through epidemiological investigation and subsequent criminal findings; the latter requires a separate evidentiary assessment of sustained technical capabilities and organizational arrangements.

This case also highlights an important problem in the relationship between public-health evidence and criminal attribution. An epidemiological investigation can establish that an unusual outbreak occurred, determine the number of identified cases and identify patterns consistent with a particular exposure, yet the determination that the event was deliberate may require information beyond ordinary disease surveillance. Investigators must distinguish what the clinical evidence establishes from what subsequent testimony, criminal proceedings or other records establish about intention and responsibility. The World Health Organization recognizes this broader challenge in its guidance on deliberate events, emphasizing that an incident may initially be detected without clear evidence concerning whether its cause was accidental or intentional. Deliberate Events — World Health Organization, April 2024. Organizzazione Mondiale della Sanità

The 2001 anthrax letters and the transformation of biological terrorism investigations

The anthrax letter attacks in the United States in 2001 established another fundamentally different historical pattern. Unlike the unsuccessful biological operations of Aum Shinrikyo or the localized Oregon outbreak, the attacks involved deliberate mailings that produced fatal disease, widespread institutional disruption and an exceptionally extensive federal investigation. According to the United States Department of Justice, five people died and 17 others became ill. On 19 February 2010, the Department of Justice, the Federal Bureau of Investigation and the United States Postal Inspection Service formally announced the conclusion of the Amerithrax investigation, identifying the late Bruce Ivins as the perpetrator in their investigative findings. Ivins died before an indictment could be secured, meaning that the government’s attribution was not tested through a completed criminal trial. Justice Department and FBI Announce Formal Conclusion of Investigation into 2001 Anthrax Attacks — US Department of Justice, February 2010. United States Department of Justice

The investigative scale illustrates the institutional consequences of a biological attack even when the number of identified disease cases is relatively limited. The Department of Justice reported more than 10,000 witness interviews across six continents, 80 searches, more than 5,750 grand-jury subpoenas and 5,730 environmental samples collected from 60 locations. These figures measure investigative activity rather than the biological severity of the incident, and they should not be used as though they were interchangeable indicators of attack effectiveness. They nevertheless demonstrate the capacity of a relatively limited number of deliberate biological exposures to generate sustained demands on law enforcement, public-health authorities, forensic laboratories and the wider government apparatus. Amerithrax or Anthrax Investigation — Federal Bureau of Investigation, historical case record. FBI

The case is particularly important because subsequent scientific review revealed the limitations of even a highly resourced investigation. In February 2011, the Department of Justice and FBI responded to an independent National Research Council assessment of the scientific methods used in the investigation. The government acknowledged that the scientific evidence alone could not definitively establish the origin of the material involved in the mailings, while maintaining that the wider investigative record supported its conclusion concerning responsibility. This distinction between forensic findings and the totality of criminal evidence is indispensable in any serious assessment of biological terrorism. Scientific analysis can provide powerful investigative leads without independently resolving every question of attribution. FBI and Justice Department Response to the National Academy of Sciences Review — US Department of Justice, February 2011. United States Department of Justice

The National Research Council’s 2011 report provides the underlying independent scientific review. Its significance extends beyond the Amerithrax case because it illustrates the institutional importance of defining precisely what particular evidence can establish, including the difference between a scientifically supported association and a legally sufficient attribution of responsibility. For biological incident preparedness, this creates a requirement for cooperation among epidemiologists, forensic scientists, criminal investigators and judicial authorities without allowing any one discipline to claim certainty beyond its evidentiary competence. Review of the Scientific Approaches Used During the FBI’s Investigation of the 2001 Anthrax Letters — National Research Council, 2011. Accademia delle Scienze

Why ideological categories cannot substitute for capability assessment

The historical record includes movements motivated by religious prophecy, political control, ideological extremism, retaliation and other objectives. These differences matter because they shape the purposes for which violence is contemplated, the relationship between perpetrators and intended victims, and the organizational context in which technical activities take place. However, ideology cannot be treated as a reliable substitute for evidence of capability. Two organizations with superficially similar beliefs may possess radically different resources, personnel arrangements and practical competencies, while actors with entirely different motivations may nevertheless produce similar public-health consequences.

The 1999 Monterey analysis found no clear and simple pattern identifying one type of organization as uniquely associated with biological incidents. Its historical observations were also affected by the presence of Aum Shinrikyo as a prominent outlier within the category of apocalyptic organizations. The concentration of a particular kind of incident around one unusually active group cannot establish that all organizations with similar ideological characteristics will behave in the same way. Moreover, the observed distribution of incidents is sensitive to reporting practices, changes in investigative attention and the inclusion or exclusion of cases that remain unsubstantiated. Historical Trends Related to Bioterrorism — Emerging Infectious Diseases, August 1999. CDC

This limitation becomes more important when attempting to assess contemporary movements. The presence of violent rhetoric, interest in biological subjects or access to scientific information should not automatically be interpreted as evidence of a functioning weapons programme. Conversely, the absence of public claims is not proof that an actor presents no threat. A defensible assessment requires independent evidence of intention, relevant actions and actual capabilities, interpreted in the context of the actor’s organization and the available investigative record.

The statistical problem: rare events, incomplete observation and incompatible datasets

Historical biological terrorism is particularly difficult to quantify because the available observations are limited, heterogeneous and subject to substantial reporting bias. A database that includes threats, hoaxes, possession, attempts and confirmed attacks may be useful for categorizing security-relevant behavior, but it cannot produce a reliable attack-success rate without a well-defined denominator and consistent criteria for inclusion. Similarly, a historical sequence of unsuccessful incidents cannot establish a stable future probability when the technological environment, population of actors and methods of detecting incidents are changing.

There is also an important distinction between the frequency of observed events and the underlying frequency of attempted events. Law-enforcement activity, improved surveillance, changes in reporting obligations and heightened public concern can all affect how many incidents become visible. A rise in reported investigations may therefore reflect changes in detection or classification as well as changes in hostile activity. The Monterey study itself noted that public reporting was incomplete and that some cases were excluded from comparison because sufficient information was unavailable. Its figures consequently provide a historical description of recorded events under specified definitions, not a complete statistical representation of all biological terrorism activity. Historical Trends Related to Bioterrorism — Emerging Infectious Diseases, August 1999. CDC

The same problem affects attempts to forecast the consequences of future technological developments. If AI changes the availability of some forms of specialist knowledge, historical data collected before those changes cannot automatically be used to estimate the resulting frequency of successful attacks. Yet the existence of technological improvements also cannot justify arbitrarily increasing historical probabilities. Such an exercise would create numerical precision unsupported by the evidence. The defensible position is to identify the assumptions underlying each historical comparison, determine which remain applicable and treat unresolved changes as analytical uncertainties requiring additional observation.

International law and the distinction between criminal conduct and biological weapons proliferation

The international legal framework also requires careful classification. The Biological Weapons Convention, opened for signature in 1972 and in force since 1975, establishes prohibitions concerning biological agents and toxins of types and quantities lacking justification for peaceful or protective purposes, alongside prohibited weapons and means of delivery. Article I is central to the Convention’s structure, while implementation depends on the legal responsibilities of States Parties and their national measures. A terrorist organization’s conduct may constitute offences under national criminal and counterterrorism legislation, but the precise applicable provisions must be determined according to jurisdiction and the conduct involved. Biological Weapons Convention, Article I — International Committee of the Red Cross, Treaty Database. ihl-databases.icrc.org

This creates an important distinction between the existence of international prohibitions and the practical ability to prevent deliberate biological incidents. Treaty commitments provide a legal framework, but implementation also requires national enforcement, competent public institutions, investigative capabilities and legitimate mechanisms for international cooperation. The history of biological terrorism demonstrates that prevention and response cannot be confined exclusively to arms-control institutions because deliberate biological incidents may initially appear as public-health emergencies. Legal classification, public-health management and criminal investigation consequently operate alongside one another rather than as substitutes.

From historical case studies to modern intelligence requirements

The historical progression from Oregon to Aum Shinrikyo and Amerithrax does not describe a single linear process in which terrorist organizations steadily became more sophisticated. Instead, the cases demonstrate different combinations of motivation, competence, organizational structure, practical consequences and investigative complexity. Oregon illustrates a deliberate biological incident with substantial localized disease consequences. Aum demonstrates the distinction between violent ambition and the successful execution of a particular unconventional weapons capability. Amerithrax demonstrates that a biological attack can produce deaths, institutional disruption and prolonged investigative demands even without representing a large-scale epidemic.

The comparative significance lies in the variety of pathways through which biological security becomes a governmental responsibility. Some cases primarily challenge the capacity to identify intentional harm within ordinary public-health surveillance. Others challenge the evaluation of terrorist intentions and technical claims. Still others impose demands on forensic attribution, judicial processes and the coordination of government institutions. A comprehensive security framework must recognize these differences rather than attempting to manage every incident according to a single hypothetical model.

For the subsequent analysis of artificial intelligence, this historical conclusion establishes an essential boundary. New technologies may change parts of the capability environment, but historical evidence does not justify presuming that all actors will benefit equally or that improvements in information access necessarily translate into successful biological operations. The appropriate analytical objective is to determine whether particular changes are supported by independent observations of relevant performance, with uncertainty preserved where such evidence is lacking.

Chapter 3 — Key judgments

The historical record demonstrates that biological terrorism comprises distinct categories of conduct rather than a uniform operational phenomenon. The confirmed consequences of the 1984 Oregon outbreak, the unsuccessful biological record of Aum Shinrikyo and the fatal 2001 anthrax mailings cannot be combined into a simple trend or a transferable attack-success probability. Their comparative value lies in clarifying the differences between intention, preparation, demonstrated effects, investigative attribution and institutional consequences.

The strongest intelligence judgment is that historical rarity and repeated failure do not constitute a guarantee of future safety, while technological advancement alone does not establish that violent actors have acquired operational biological capabilities. Both propositions require careful separation from claims that have not been independently demonstrated.

What would change this assessment: A validated, consistently defined and sufficiently comprehensive international incident dataset, together with corroborated evidence showing how changes in relevant technology and organizational arrangements affect observed capabilities over time.

Open official record: The main unresolved requirements are internationally comparable incident classifications, systematic treatment of unconfirmed cases, reliable distinctions between intended and observed consequences, and sufficiently detailed information concerning the interaction between public-health investigations and criminal attribution.

Pillar I Interactive Historical Intelligence Scheme and Evidence Table

The historical figures, classifications and interpretations are supported by the CDC-hosted historical incident study, the Oregon outbreak investigation, the FBI’s Amerithrax case record and the National Research Council’s independent forensic review.

PILLAR I · HISTORICAL INTELLIGENCE

Biological Terrorism: Evidence, Outcomes and Constraints

A historical comparison of documented incidents and the limitations of conclusions drawn from them.

1984 · OREGON
751
Identified illness cases in the historical outbreak.
1990s · AUM SHINRIKYO
10+
Reported unsuccessful biological attack attempts.
2001 · AMERITHRAX
22
Identified anthrax cases, including five deaths.

Historical Case Comparison

Oregon, 1984

The documented outbreak caused 751 illnesses. It demonstrates localized biological harm, not a universal model of terrorist capability.

Historical Evidence Matrix

The figures represent different outcomes and must not be compared as equivalent attack rates.

Case Established figure Evidence category Analytical limitation
Oregon, 1984751 illnesses Documented outbreak Not a general attack-success rate
Aum ShinrikyoAt least 10 failed attempts Historical review Not proof of permanent technical barriers
Amerithrax, 200122 cases; 5 deaths Federal investigation Attribution and scientific evidence have distinct limits
Monterey database415 recorded CBRN incidents Historical open-source compilation Incomplete reporting and heterogeneous categories

Evidence-to-Judgment Framework

An analytical classification, not an operational process.

01 Stated intention
02 Documented resources
03 Demonstrated performance
04 Confidence and uncertainty
These categories must be evaluated independently. Evidence of intention or resources is not proof of demonstrated capability. Intelligence confidence depends on corroboration, provenance and the limits of the available record.
Source basis: CDC-hosted historical analysis (1999); Oregon outbreak epidemiology (1997); FBI and US Department of Justice Amerithrax records (2001–2011). Historical figures are not live data. No risk probabilities are calculated.

OPEN-SOURCE ANALYTICAL ASSESSMENT | 21 NOVEMBER 2026

PILLAR II — ARTIFICIAL INTELLIGENCE AND THE TRANSFORMATION OF BIOLOGICAL RISK

Frontier AI capabilities, organizational adaptation, dual-use biotechnology, model governance and the changing architecture of biological security

Chapter 4. Frontier AI, Biological Knowledge and the Changing Boundaries of Technical Expertise

The emergence of a new scientific capability environment

Artificial intelligence is changing the relationship between access to scientific information, the interpretation of specialized knowledge and the practical organization of biological research. The most consequential development is not simply that language models can retrieve or reproduce scientific material. It is that increasingly capable systems can integrate information across disciplines, reason about technical questions, assist with scientific interpretation and support researchers engaged in legitimate laboratory activities. For national security institutions, this creates an important analytical challenge: capabilities that accelerate medicine, diagnostics, public-health research and biotechnology development may also change some of the informational constraints historically considered relevant to the prevention of biological weapons proliferation.

The distinction between scientific knowledge and operational competence remains fundamental. An AI model may answer complex biological questions accurately without possessing the practical judgment, physical access, experimental reliability or institutional accountability required to conduct real-world research. Similarly, an impressive performance on a scientific examination cannot establish that the same model will perform reliably when confronted with incomplete observations, equipment failures, inconsistent data or situations that demand decisions beyond its evaluated capabilities. These distinctions become particularly important when AI performance is described in terms of human qualifications, such as doctoral-level expertise, because a researcher’s professional competence encompasses substantially more than the ability to answer technical questions.

The United Kingdom’s AI Security Institute provides one of the most important public institutional baselines for assessing this transition. Its Frontier AI Trends Report, based on evaluations conducted since November 2023, documents substantial improvements across scientific and other security-relevant capabilities. The Institute reports that frontier models surpassed human expert baselines on selected chemistry and biology assessments, while laboratory-support evaluations demonstrated further advances. The report also identifies weaknesses in model safeguards, indicating that improvements in capability and improvements in protective controls do not necessarily proceed at the same rate. These results establish a changing technological environment, but they do not establish a corresponding increase in the real-world success rate of biological terrorism. Frontier AI Trends Report — UK AI Security Institute, 2025. AI Security Institute

Scientific benchmarks: what the available measurements actually establish

The AISI findings are significant partly because they provide numerical information about measured scientific performance. The Institute’s public account describes evaluations involving more than 30 advanced AI models across its wider testing programme and reports expert baseline scores of 38% for Biology QA and 48% for Chemistry QA in the relevant evaluation series. These figures are not estimates of the percentage of all biological knowledge possessed by a human scientist, nor do they measure the probability that an AI system could support an illicit biological operation. They are performance baselines on particular evaluation sets, established using qualified participants under the testing arrangements adopted by the Institute.

The Institute reports that the first evaluated model exceeded its biology expert baseline in 2024, with subsequent frontier models demonstrating substantially stronger performance. The result matters because it demonstrates the inadequacy of assuming that difficult scientific questions necessarily remain beyond general-purpose AI. Nevertheless, it would be incorrect to translate these scores into a claim that AI has surpassed the overall professional competence of biomedical researchers. A benchmark can evaluate performance on a defined collection of questions while leaving substantial uncertainty concerning experimental execution, independent scientific judgment, reliability over extended projects and performance under changing physical conditions. Five Key Findings from Our First Frontier AI Trends Report — UK AI Security Institute, 2025. AISI Work

The analytical importance of these distinctions becomes apparent when considering the concept of expertise itself. A human specialist does not simply hold a collection of factual answers. Professional expertise includes recognizing the limits of available information, identifying unexpected observations, questioning inappropriate assumptions, evaluating alternative interpretations and knowing when specialist consultation is required. Some of these capabilities can be measured in controlled exercises, while others emerge through practical experience and professional accountability. An AI system may perform strongly on one group of tasks and poorly on another, creating an uneven capability profile that a single headline score cannot adequately represent.

For biological security assessment, the relevant question is therefore not whether a model is generally more or less knowledgeable than a scientist. It is whether its assistance produces an observable improvement in a specific, appropriately defined task, compared with a realistic alternative such as access to conventional reference material, standard software or qualified human guidance. Such evaluations should also distinguish between users with different levels of prior competence. An improvement observed among trained researchers cannot automatically be assumed to apply to novices, while a lack of improvement among novices does not prove that the technology has no value to established scientific organizations.

Laboratory assistance and the difference between conceptual accuracy and experimental reliability

Laboratory-support performance introduces a second category of evidence beyond scientific question-answering. The AISI reports progress in evaluations involving the generation and assessment of experimental guidance and laboratory troubleshooting. Its broader findings state that some AI-generated scientific guidance has been judged accurate and shown to be feasible in laboratory settings. These results demonstrate that AI can contribute more than the reproduction of textbook knowledge, but they do not imply that frontier models can reliably conduct autonomous biological research or that they can overcome every practical constraint relevant to dangerous biological applications. Frontier AI Trends Report — UK AI Security Institute, 2025, chemistry and biology findings. AI Security Institute

The distinction between scientific guidance and reproducible experimental success is especially important because laboratories are environments in which apparently small uncertainties can have substantial consequences. Researchers must interpret observations that may be incomplete, recognize inconsistencies between expectations and results, maintain appropriate controls and determine whether findings satisfy the standards of the relevant discipline. An AI system that produces plausible explanations may assist this process, but plausibility is not evidence of correctness. Inaccurate or overly confident recommendations may introduce additional uncertainty rather than resolve it.

Evaluating AI-supported laboratory research therefore requires more than asking whether a model can produce an answer judged reasonable by a specialist. Stronger evidence comes from appropriately supervised studies examining whether assistance improves the quality, consistency or efficiency of legitimate research activities under real working conditions. These studies should account for participant qualifications, task difficulty, access to alternative resources, reproducibility, measurement uncertainty and the possibility that model-generated outputs introduce new errors. Any security-relevant interpretation must remain proportionate to what those studies actually measure.

The UK AISI’s findings provide grounds for sustained monitoring because they demonstrate improvement in selected capabilities that were previously less developed. They do not, however, justify a general conclusion that practical biological barriers have disappeared. National security institutions should resist both simplistic interpretations: the claim that scientific expertise remains completely inaccessible to AI systems, and the opposite claim that strong benchmark performance is sufficient to establish a direct pathway to an effective biological weapon.

The meaning of measurable technological acceleration

The pace of improvement is relevant to security planning because institutional safeguards often develop through processes that are slower than technological releases. Governments establish regulatory frameworks, scientific advisory arrangements, procurement systems and international agreements through procedures involving consultation, legal review and administrative implementation. AI developers may update models and deployment architectures much more frequently. This difference creates the possibility that an evaluation conducted for one generation of systems will become less informative about systems available later.

Nevertheless, acceleration must be defined carefully. A numerical improvement on a benchmark cannot be assumed to represent the same percentage improvement in laboratory productivity, scientific discovery or biological misuse capability. Different evaluations measure different competencies, and performance improvements can vary considerably according to the task, model configuration and surrounding software environment. Changes in the composition or difficulty of an evaluation can also affect apparent trends. A measured capability trajectory should therefore be reported alongside its evaluation method and interpreted only within that scope.

This matters because strategic forecasts are particularly vulnerable to misleading extrapolation. A model that improves rapidly on scientific questions does not necessarily improve at an equivalent rate in physical-world reliability. Equally, the existence of demanding practical constraints today does not establish that their importance will remain unchanged. The appropriate institutional response is to maintain a regularly updated evaluation programme with stable reference tasks, transparent definitions, independent review where feasible and explicit separation between observed performance and hypothetical future consequences.

The difference between general-purpose AI and specialized biological AI

The biological AI ecosystem is not composed exclusively of conversational language models. It includes systems developed for specialized scientific activities, such as the interpretation of biological data, protein-related research, medical analysis and computational modelling. General-purpose systems and specialized scientific tools can support different parts of legitimate research, but their capabilities, limitations and governance arrangements are not interchangeable.

A general-purpose assistant is designed to respond to a wide range of instructions, while a specialized scientific system may be optimized for a narrower class of tasks. The existence of specialized tools complicates broad claims about the capabilities or limitations of AI as a single category. It also raises questions concerning the integration of different technologies within research institutions, where information systems, scientific databases and computational tools may be combined under human oversight.

For security governance, this means that evaluating only the outputs of a single general-purpose chatbot may leave important questions unanswered. However, it would also be inappropriate to infer that combining several tools automatically produces a reliable autonomous scientific capability. Integration introduces its own requirements for compatibility, validation, data quality, professional oversight and accountability. Errors originating in one system can propagate into others, and the resulting output may appear more authoritative without being more accurate.

The relevant governance issue is whether safety assessments adequately reflect the complete system that authorized users actually employ, while preserving legitimate scientific access. This requires attention to software environments, institutional controls, access arrangements and the quality of scientific verification, rather than treating every model as an isolated product.

Provider precautionary measures and the evidence they do not establish

The major AI developers have publicly acknowledged that advances in scientific capability require specific biological safeguards. On 18 June 2025, OpenAI published its approach to preparing for future AI capabilities in biology, including specialist evaluation, collaboration with outside experts, adversarial testing, deployment controls and monitoring for misuse. The company described biological security as an area requiring precautionary preparation before a severe incident occurs and emphasized the continued relevance of physical access to laboratories and sensitive materials. These statements document the developer’s risk-management approach and expectations; they do not independently establish the probability of a particular future biological attack. Preparing for Future AI Capabilities in Biology — OpenAI, June 2025. OpenAI

Anthropic’s decision to activate its AI Safety Level 3 deployment and security protections in connection with Claude Opus 4 on 22 May 2025 provides another important institutional example. Its announcement distinguished safeguards intended to reduce prohibited model use from security controls designed to protect model parameters. Crucially, the company stated at the time that the additional precautions were provisional and that it had not definitively established that the model had crossed the relevant capability threshold. The distinction between precautionary action and confirmed dangerous capability is essential. A protective measure may be justified by uncertainty and the potential consequences of misuse without constituting evidence that an identified threat has already materialized. Activating AI Safety Level 3 Protections — Anthropic, May 2025. Anthropic

These initiatives demonstrate a growing recognition among developers that biological capability assessment cannot be treated as an ordinary product-performance evaluation. They also expose the limits of relying exclusively on developer declarations. Companies may possess extensive internal evaluation data and direct knowledge of their systems, but their public reporting can differ in scope, methodology and completeness. Government authorities and qualified independent evaluators therefore require sufficient access to assess the strength of relevant evidence, the validity of safeguards and the consequences of different deployment arrangements.

The problem of translating scientific capability into national security risk

The most consequential analytical distinction in this chapter is between measured capability, practical assistance and demonstrated security consequences. These categories are connected, but the evidence required to establish each is different. An AI system may demonstrate specialist knowledge in an evaluation, contribute to legitimate scientific work under controlled conditions and still have an uncertain effect on malicious actors who lack relevant practical resources or experience. Conversely, the absence of a demonstrated real-world security consequence cannot establish that the underlying technology presents no future risk.

For government decision-making, this uncertainty should be expressed explicitly rather than converted into an invented probability. Models can be evaluated against defined scientific tasks; legitimate laboratory assistance can be studied under appropriate supervision; safeguards can be tested; and changes in deployment arrangements can be documented. The resulting evidence can support increasingly precise judgments about particular capabilities. It does not automatically support numerical forecasts of the frequency, severity or success of future terrorist operations.

The broader finding is that biological knowledge is becoming more accessible through advanced computational systems, while the translation of that knowledge into reliable practical capability remains incompletely understood. This is a sufficient basis for updating biological security evaluations and preparedness arrangements, without relying on speculative claims concerning operational breakthroughs.

Chapter 4 — Key judgments

The scientific capabilities of frontier AI systems are advancing on several measured dimensions relevant to biological research. Government evaluations establish that selected models can exceed expert reference performance on specific scientific assessments and provide increasingly capable laboratory-support assistance. These findings are significant because they challenge assumptions about the exclusivity of specialized scientific knowledge, but they do not demonstrate that AI can independently reproduce the full competence of professional researchers or reliably enable biological weapons development.

The policy consequence is the need for task-specific, independently assessable and periodically updated evaluations, distinguishing knowledge performance from practical laboratory outcomes and actual security consequences.

What would change the assessment: Reproducible evidence demonstrating consistent improvement or persistent failure in legitimate practical research settings, measured against appropriate non-AI alternatives and supported by transparent methods.

Open official record: Comparative information concerning sustained real-world research performance, differences between user skill levels and the robustness of scientific evaluations remains incomplete.

Chapter 5. Organizational AI, Decision-Making, Knowledge Transfer and Adaptive Threat Networks

Artificial intelligence as an institutional system rather than an isolated technical assistant

The organizational consequences of artificial intelligence extend beyond the scientific capabilities of individual models. The more fundamental institutional change concerns the ability to embed computational systems within established processes for collecting information, interpreting reports, maintaining organizational records, coordinating personnel and supporting decisions. In legitimate government departments, research laboratories and private enterprises, these functions are associated with improving administrative consistency, reducing information fragmentation and making specialized knowledge accessible across organizational boundaries. The security relevance arises when similar technologies become available to organizations whose objectives conflict with public safety and the law.

This creates a different analytical problem from the measurement of scientific expertise. A model may perform well on a technical evaluation while providing limited value to an organization that cannot integrate its outputs into responsible decision-making. Alternatively, a relatively modest system may have substantial administrative effects when used within an institution already possessing qualified personnel, reliable procedures and established means of verifying information. The relationship between AI capability and organizational performance is therefore neither automatic nor necessarily proportional. It depends on the interaction between technical systems, users, institutional incentives and the quality of existing governance.

The United States National Institute of Standards and Technology provides a useful framework for examining these interactions without assuming that technological sophistication necessarily produces beneficial outcomes. Its AI Risk Management Framework: Generative Artificial Intelligence Profile, published on 26 July 2024, describes the need to incorporate AI-related risks into organizational governance, mapping, measurement and management. Although intended for responsible AI deployment rather than counterterrorism analysis, the framework establishes that risks arise across an entire system’s lifecycle and cannot be adequately assessed by evaluating the model’s technical performance alone. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — NIST, July 2024. NIST

Institutional memory, information continuity and the limits of automated interpretation

One of the most important effects of digital systems on organizations is their capacity to preserve information beyond the immediate availability of individual personnel. Conventional organizational records already perform this function, but AI systems can provide additional means of searching, summarizing and interpreting extensive collections of material. In government and scientific institutions, these applications may reduce the administrative burden of locating relevant documents, reviewing earlier decisions or identifying inconsistencies between different reports. Their value depends on the reliability of the underlying information, the system’s treatment of uncertainty and the existence of qualified human review.

The security implications are complex. An organization may become more efficient at retrieving information without becoming more accurate in its conclusions. If an AI system repeatedly summarizes incomplete or misleading records, it can reproduce the same errors while creating an impression of institutional coherence. When model outputs are incorporated into later reports, the organization may develop a cycle in which unverified claims gradually acquire apparent authority through repetition. This is particularly problematic in environments characterized by weak accountability or strong pressure to produce favorable assessments.

Consequently, the existence of persistent AI-supported records should not be confused with demonstrable organizational learning. Genuine learning requires a distinction between observations, interpretations and decisions, together with a mechanism for correcting information that later proves inaccurate. An organization may preserve more information than before while remaining unable to determine which information is reliable. The effect of AI therefore depends on institutional discipline as much as computational performance.

Algorithmic authority and the redistribution of decision-making power

AI systems can also change the perceived distribution of expertise inside an institution. Where personnel lack confidence in their own technical knowledge, recommendations produced by a sophisticated computational system may appear authoritative even when the system’s reliability has not been demonstrated in the relevant context. This phenomenon creates a governance problem because the apparent objectivity of an automated output can obscure the assumptions, errors and limitations involved in producing it.

The consequences are not limited to situations in which leadership delegates formal authority to AI. Even when human personnel retain nominal control, their decisions may become heavily influenced by model-generated summaries, evaluations or recommendations. If organizational culture discourages challenging established conclusions, the presence of AI may reinforce rather than correct existing deficiencies. A system that appears to provide rapid access to specialist judgment can become an additional source of institutional error when its users treat uncertainty as resolved simply because an answer has been generated.

For security analysis, it is therefore insufficient to ask whether a violent organization possesses AI systems or reports using them. The more important issue is whether credible evidence demonstrates that the technology has altered actual organizational decisions and whether the resulting decisions were better informed. This distinction protects against the tendency to interpret technological adoption as an automatic increase in operational effectiveness.

The NIST framework is relevant because it treats governance as a continuous responsibility involving documentation, evaluation, appropriate human oversight and the identification of risks throughout the lifecycle of AI systems. Its principles can inform institutional assessments of automation bias and accountability, although they do not provide empirical evidence about the internal practices of particular extremist organizations. NIST AI 600-1 — National Institute of Standards and Technology, 2024. NIST

Knowledge transfer and the challenge of validating AI-generated information

Organizations face a recurring difficulty when expertise is concentrated among relatively few individuals while decisions involve many departments or personnel. AI-assisted information systems can potentially support communication across those boundaries, particularly when users need to interpret unfamiliar terminology, organize technical documents or identify questions requiring specialist review. These are important functions in legitimate interdisciplinary research and public administration, where technical information often must be translated into decisions involving finance, personnel, legal requirements and operational constraints.

However, the existence of an AI-mediated communication channel does not establish that information becomes more accurate as it moves between users. Technical qualifications remain relevant because the recipient of a recommendation may be unable to determine whether the information is incomplete, misleading or inappropriate for the circumstances. Moreover, information originating from different documents may contain conflicting assumptions, and a model-generated synthesis may conceal those conflicts rather than resolve them. Effective knowledge transfer therefore requires verification procedures, not merely faster communication.

The distinction has implications for the assessment of emerging threat networks. An investigator may encounter evidence of AI-generated material or AI-supported communication, but that evidence alone cannot establish the reliability of the information, the competence of the users or any resulting change in practical capability. The proper evidentiary question concerns what can be corroborated about organizational behavior and consequences, rather than what a model might theoretically have helped the organization accomplish.

The reliability problem in adaptive organizations

The concept of an adaptive organization is frequently associated with the ability to respond to new information and modify behavior. In practice, adaptation is not necessarily beneficial, effective or even rational. Organizations can change their procedures in response to inaccurate observations, adopt inappropriate recommendations or reinforce unsuccessful policies through increased commitment. AI may accelerate the processing of information without improving the quality of the underlying judgment.

An important analytical distinction therefore exists between adaptation and improvement. Adaptation can be observed through changes in organizational procedures, personnel responsibilities or technology use. Improvement requires evidence that those changes produced better outcomes according to a clearly defined measure. In counterterrorism, these observations may be difficult to obtain because much of the relevant evidence is confidential, incomplete or reconstructed through investigations after the event.

This uncertainty is consequential. If security authorities attribute an organization’s improved performance to AI without sufficient evidence, they risk misunderstanding the actual drivers of change. Changes in membership, access to resources, external relationships, organizational priorities or the surrounding environment may provide alternative explanations. Conversely, rejecting a potential AI contribution simply because other factors are present would be equally unjustified. Sound analysis requires evidence that distinguishes the different possible explanations rather than selecting the most technologically novel one.

AI-enabled information operations and the expansion of organizational activity

A wider security concern arises from the use of generative AI in communication, document production, translation and public-facing information activities. These applications differ fundamentally from biological weapons development, but they may affect the visibility, messaging and administrative functioning of extremist movements. General-purpose models can support legitimate multilingual communication and content preparation, and those capabilities may also be misused for propaganda or other harmful purposes.

The relevance to biological security is indirect. A group that uses AI for communication does not thereby acquire specialist biological competence, just as a group that automates administrative tasks does not necessarily acquire an effective technical programme. However, greater digital sophistication may complicate the interpretation of organizational material, particularly where AI-generated content creates uncertainty about authenticity, authorship or the relationship between public statements and actual organizational capabilities.

For government institutions, this reinforces the importance of evidence provenance. Generated documents, purported technical assessments and organizational announcements should be evaluated according to their origin and corroboration, rather than the sophistication of their presentation. The ease of producing convincing digital material increases the importance of distinguishing genuine activity from claims designed to influence observers.

Implications for intelligence collection and institutional assessment

The principal government challenge is to evaluate organizational AI without turning its mere presence into a threat indicator. Legitimate AI adoption is widespread, and the possession or use of general-purpose systems is not evidence of criminal intent. An assessment that treats AI access as inherently suspicious would produce extensive false positives, undermine legitimate research and misallocate investigative resources. The appropriate threshold for security attention must therefore depend on lawful, corroborated indicators of harmful intent or conduct, interpreted within the relevant legal framework.

This implies a distinction between three categories of observation. The first concerns access to technology, which is common and generally lawful. The second concerns organizational integration, which may be relevant to understanding how an institution functions but does not establish harmful capability. The third concerns independently demonstrated security-relevant conduct, which requires its own evidentiary basis. Treating these categories separately permits more proportionate and reliable assessments.

The international dimension makes this discipline especially important because governments operate under different legal regimes governing intelligence collection, privacy, scientific research and the sharing of information with technology providers. Measures intended to improve situational awareness must be assessed against applicable law and the need to avoid discrimination, overcollection or unjustified interference with legitimate activity. Technological monitoring alone cannot replace competent human analysis or established evidentiary standards.

Chapter 5 — Key judgments

AI can change how organizations collect, preserve, interpret and circulate information, but greater information-processing capacity does not automatically produce better decisions. Institutional incentives, user competence, verification procedures and human accountability continue to determine whether automated assistance contributes to reliable organizational performance. In security analysis, AI adoption must be distinguished from both malicious intent and demonstrated operational capability.

The central intelligence requirement is to assess observable organizational consequences, while preserving alternative explanations and avoiding the assumption that every technological change produces a measurable increase in threat.

What would change the assessment: Independently corroborated evidence showing that AI integration produces repeatable organizational performance changes in relevant settings, with appropriate attention to competing explanations.

Open official record: Publicly verifiable information concerning the actual performance effects of AI adoption within clandestine violent networks remains limited. Reported use, perceived usefulness and demonstrated operational consequences must not be conflated.

Chapter 6. The Dual-Use Biotechnology Ecosystem: Open Models, Commercial Services, Governance and Security Exposure

Biological security within an increasingly distributed technology ecosystem

The contemporary biological research environment is shaped by an extensive combination of academic institutions, commercial biotechnology companies, scientific software providers, AI developers, cloud infrastructure, specialized equipment suppliers and public research agencies. These institutions contribute to advances in medicine, diagnostics, agriculture, public health and industrial biotechnology. Their activities also create a governance challenge because scientific capabilities may have legitimate and potentially harmful applications, and responsibility for safety is distributed across organizations operating under different national laws, professional standards and commercial incentives.

The World Health Organization addressed this issue in its September 2022 Global Guidance Framework for the Responsible Use of the Life Sciences. The framework emphasizes that biological risk governance is a shared responsibility involving governments, research institutions, scientists, funders, publishers, educators, private companies and other participants in the research lifecycle. This is an important departure from approaches that treat biological security primarily as a problem of physical laboratory protection or national criminal enforcement. Modern biological security also concerns how research is proposed, assessed, financed, communicated, published and used. Global Guidance Framework for the Responsible Use of the Life Sciences: Mitigating Biorisks and Governing Dual-Use Research — World Health Organization, September 2022. Organizzazione Mondiale della Sanità

Artificial intelligence adds a new governance layer because research institutions may use computational systems supplied by organizations outside the traditional biological sciences. Responsibility can consequently become fragmented between the developer of a foundation model, the provider of a scientific application, the institution operating the system and the individuals responsible for the biological research. Each may possess only partial visibility into how the complete system functions. This creates challenges for accountability, particularly when a model is adapted, integrated into additional software or used in circumstances different from those originally evaluated.

Closed-weight, hosted and open-weight models: distinct security arrangements

The distinction between hosted and open-weight AI models is fundamental to understanding present governance limitations. In centrally operated services, the provider generally retains substantial control over model deployment, account access and the surrounding safeguards. Depending on the architecture and applicable policies, the provider may be able to update protective systems, restrict particular uses, monitor certain forms of misuse and disable access in response to identified violations. These capabilities do not guarantee the prevention of harmful activity, but they provide potential intervention mechanisms that are unavailable once a model has been widely redistributed outside the provider’s infrastructure.

Open-weight systems operate under different conditions. When model parameters become available for independent deployment, organizations and individuals may be able to operate modified versions without continuing access to the original developer’s servers. This supports legitimate scientific independence, local processing, transparency, experimentation and competition. It also limits the developer’s ability to impose post-release changes or ensure that downstream users retain the originally supplied safeguards.

The UK AI Security Institute’s analysis of open-weight model risk management explicitly recognizes these competing considerations. It identifies advantages for open research and innovation while explaining that safeguards may be modified or removed after a model is distributed. The Institute considers measures including evaluation prior to release, documentation, training-data governance, staged deployment and assessments that take account of possible downstream modification. It also emphasizes that existing techniques do not provide absolute guarantees of safety. Managing Risks from Increasingly Capable Open-Weight AI Systems — UK AI Security Institute. AISI Work

The policy consequence is that a security strategy designed solely around refusing dangerous prompts through centrally hosted services will not address every deployment arrangement. Equally, it would be misleading to characterize all open-weight systems as inherently unsafe or to conclude that every closed-weight system is adequately protected. The relevant considerations include model capabilities, release arrangements, the robustness of protective measures, the possibility of independent modification and the consequences of limiting access to scientific tools.

Model safeguards and the meaning of defensive effectiveness

AI safeguards comprise multiple categories of intervention. Some are intended to reduce the generation of harmful outputs; others concern access management, system security, abuse detection, independent evaluation or the governance of model releases. Their effectiveness should be examined individually, because a measure that performs well under one testing arrangement may be less effective under another. The presence of multiple safeguards can provide important additional protection, but it does not establish that misuse is impossible.

The AISI’s frontier-model findings are particularly relevant because they indicate that safeguard performance varies considerably across systems and categories of misuse. The Institute reported that stronger safeguards increased the effort required to circumvent some protections during controlled testing, while vulnerabilities remained identifiable in the systems evaluated. The report also found that general model capability was not a reliable indicator of safeguard strength, reinforcing the need to evaluate security performance independently rather than infer it from the sophistication of the underlying model. Frontier AI Trends Report — UK AI Security Institute, 2025. AI Security Institute

This evidence supports a more demanding interpretation of safeguard effectiveness. A system should not be considered adequately protected merely because it refuses a small collection of prohibited test requests. Nor does the identification of a vulnerability necessarily establish that harmful real-world consequences are likely. Evaluation must distinguish the existence of a technical weakness from its severity, reproducibility and practical significance. These distinctions are particularly important when researchers and developers communicate results to policymakers, because imprecise descriptions can either exaggerate the danger or obscure vulnerabilities requiring attention.

The need for evaluation also introduces a governance problem concerning who has access to test results, how sensitive findings are handled and how independent assessments can be conducted without unnecessarily disseminating harmful information. Governments and developers may need mechanisms for confidential reporting, qualified independent review and responsible remediation. Such arrangements should aim to improve accountability while respecting legitimate research access and applicable legal protections.

Commercial biotechnology and the distribution of responsibility

The commercial biotechnology sector contributes substantially to legitimate scientific innovation through services, equipment, software and research infrastructure. The existence of commercial providers can support responsible research by giving qualified institutions access to standardized products and professionally managed services. At the same time, the diversity of providers and jurisdictions complicates the development of consistent security expectations.

The policy issue is not simply whether a particular organization possesses appropriate internal controls, but whether responsibilities remain clear when research activities involve multiple service providers. A legitimate project may rely on external computational systems, data infrastructure, commercial research services and university oversight arrangements. If each organization assumes that another party is responsible for assessing relevant risks, accountability can become fragmented. Conversely, imposing indiscriminate restrictions on all scientific services could create serious costs for public health, academic research and biotechnology competitiveness.

The WHO’s 2022 framework is relevant because it addresses biological risk throughout the research lifecycle rather than treating governance as a single checkpoint. Its approach provides a basis for considering institutional responsibilities in research oversight, education, funding, scientific communication and the private sector. The framework does not establish that all participating institutions have implemented equivalent safeguards, and it does not replace national law. It instead provides principles for developing governance suited to different institutional circumstances. Global Guidance Framework for the Responsible Use of the Life Sciences — WHO, September 2022. Organizzazione Mondiale della Sanità

Model security, scientific confidentiality and competing public interests

The security of advanced AI systems also includes the protection of model infrastructure and proprietary technical assets. Providers may regard the unauthorized acquisition of model parameters as a security concern because it can compromise commercial interests and reduce control over deployment. In systems with capabilities relevant to biological misuse, such incidents may also have implications for the availability of safeguards. The strength and implementation of model-security controls therefore form part of the broader governance environment.

Anthropic’s May 2025 announcement is illustrative because it distinguished security measures protecting model weights from deployment measures intended to reduce harmful use. These functions address different problems and should not be treated as equivalent. Preventing unauthorized access to model infrastructure does not automatically ensure that authorized outputs are safe, while controlling model responses does not by itself protect infrastructure against compromise. Activating AI Safety Level 3 Protections — Anthropic, May 2025. Anthropic

At the same time, protecting model security must be considered alongside legitimate interests in scientific transparency, reproducibility, research independence and the ability of institutions to process sensitive scientific information within controlled environments. Open-weight deployment may provide important advantages in circumstances where researchers require local systems or greater technical transparency. A governance framework that ignores these benefits may encourage unnecessary concentration of technological control and impose costs on legitimate scientific development without demonstrating proportionate security improvements.

The resulting policy challenge is not to select one universal distribution model, but to establish decision criteria appropriate to the capabilities, risks and intended uses of different systems. These criteria should remain adaptable as technologies change and should be subject to independent scrutiny where practicable.

European preparedness and the industrial dimension of biological resilience

The European Union provides an important example of biological preparedness extending beyond the governance of AI models. Through the Health Emergency Preparedness and Response Authority, the European Commission coordinates activities involving threat assessment, medical countermeasures, manufacturing capacity and health emergency readiness. These responsibilities are relevant to biological security because the consequences of a deliberate biological incident may require rapid action by institutions that were established primarily for public-health preparedness rather than counterterrorism.

The Commission’s HERA documentation identifies three broad priority threat categories established in 2022: pathogens with high pandemic potential, chemical, biological, radiological and nuclear threats, and antimicrobial resistance. It also describes efforts to improve epidemic intelligence, laboratory and diagnostic capacities, sequencing and wastewater surveillance. In its published account of sequencing-capacity support, the Commission identifies €39 million in direct grants to Member States, intended principally to strengthen genomic surveillance capabilities. This is a specific funding commitment within a broader preparedness architecture, not a total European biological security budget or a measure of the Union’s capability to prevent deliberate attacks. Threat Assessment and Intelligence Gathering — European Commission, HERA. Public Health

The policy relevance lies in the relationship between detection, medical response and industrial preparedness. Biological incidents may create simultaneous demands for diagnostics, protective equipment, specialized medical products, laboratory capacity, secure information exchange and public communication. These demands are not eliminated by AI safeguards, because model-level interventions address only one component of the wider security environment. The ability to respond to a biological emergency therefore depends on institutional capacities that must be maintained independently of assumptions concerning the identity or technology of a potential perpetrator.

The Commission’s preparedness framework also reflects the importance of medical countermeasure supply chains and industrial capacity. A national or European response may depend on products manufactured across different jurisdictions, requiring coordination between health authorities, procurement bodies, manufacturers and international partners. The capacity to identify demand does not automatically establish that adequate supplies can be produced or distributed within the required time. These distinctions between planning, procurement, stockpiling, manufacturing capacity and actual availability will be examined in the country-specific assessments of Pillar III rather than inferred from the existence of European programmes.

The governance problem created by fragmented jurisdictions

AI and biotechnology are international sectors, while regulatory authority remains divided among national governments and regional institutions. A model may be developed in one jurisdiction, hosted in another, adapted by a third-party organization elsewhere and used within a scientific institution operating under a different regulatory framework. The relevant obligations may arise from overlapping rules concerning AI, biotechnology, scientific research, data protection, cybersecurity, export controls and criminal law.

This fragmentation can complicate accountability even when every participating institution operates lawfully. Requirements may differ in scope, terminology, enforcement procedures and the treatment of research exemptions. A declaration that an organization complies with a particular AI regulation does not establish compliance with all relevant biological security requirements. Similarly, the existence of scientific research oversight does not necessarily demonstrate that model-level misuse risks have been evaluated.

International cooperation is therefore necessary, but its effectiveness depends on the ability to translate broad principles into workable institutional arrangements. Technical evaluation standards, appropriate information-sharing mechanisms, professional oversight, responsible model-release procedures and compatible emergency-response practices may all contribute to a more coherent approach. However, each requires attention to legal authority, implementation costs and possible consequences for legitimate scientific activity.

From technology governance to integrated biological security

The principal finding of this chapter is that biological security cannot be organized around a single technological intervention. AI developers can influence model design, deployment and certain forms of misuse detection. Scientific institutions can establish research oversight and verification practices. Commercial providers can implement responsibilities appropriate to their services. Public-health authorities can strengthen surveillance and response capacity. Governments can coordinate legal obligations, intelligence, emergency preparedness and international cooperation. None of these functions fully substitutes for the others.

The WHO’s framework offers a general institutional basis for shared responsibility, while the NIST approach offers methods for evaluating AI-related risks throughout the system lifecycle. The AISI provides evidence concerning frontier capabilities and the limitations of current safeguards. HERA illustrates the wider public-health and industrial preparedness environment. Taken together, these records support a layered and adaptive approach to biological security, while leaving substantial uncertainty concerning the effects of AI on the operational capabilities of violent non-state actors. WHO Global Guidance Framework, September 2022; NIST AI 600-1, July 2024; UK AISI Frontier AI Trends Report, 2025; European Commission HERA Preparedness Framework. Organizzazione Mondiale della Sanità+3

Chapter 6 — Key judgments

The dual-use biotechnology environment increasingly depends on technologies and services governed by different institutions, commercial arrangements and legal jurisdictions. Hosted and open-weight AI systems create distinct opportunities and limitations for safeguards. No single class of model, provider control or regulatory instrument can independently guarantee biological security.

The strategic requirement is to establish defence in depth across the technological and institutional ecosystem, while ensuring that measures intended to reduce misuse do not unnecessarily restrict legitimate biomedical research, scientific transparency or public-health innovation.

What would change the assessment: Independently verified improvements in safeguard robustness, reproducible evidence concerning the consequences of different release arrangements, or new binding international measures that materially alter the responsibilities of AI developers and biological research institutions.

Open official record: Comparable evidence concerning safeguard effectiveness across deployment arrangements, the implementation of biological research governance and the practical interoperability of national and international security institutions remains incomplete.

Pillar II Interactive Intelligence Matrix and 3D Governance Scheme

The evidence derives from the UK AI Security Institute’s Frontier AI Trends Report, AISI’s open-weight governance assessment, NIST’s Generative AI Profile, the WHO’s 2022 life-sciences governance framework and the European Commission’s HERA programme documentation.

PILLAR II · TECHNOLOGY AND GOVERNANCE

AI and Biological Security: Evidence and Governance

Measured scientific capabilities, organizational uncertainty and institutional safeguards.

BIOLOGY QA · AISI BASELINE
38%
Human expert reference score on a selected assessment.
CHEMISTRY QA · AISI BASELINE
48%
Human expert reference score on a separate assessment.
HERA · SEQUENCING SUPPORT
€39m
Direct grants for Member State sequencing capacity.

Reference Performance on Selected Scientific Evaluations

These are human expert reference percentages on two different AISI tests. They are not biological attack probabilities or measures of laboratory productivity.

WHAT THE GRAPH MEANS

The values identify reference performance by qualified human participants. AISI reports that later frontier models surpassed these reference levels on the relevant evaluations. This chart does not display a current AI model score because no comparable single value has been specified for one model and date.


OPEN-SOURCE ANALYTICAL ASSESSMENT | 21 NOVEMBER 2026

PILLAR III — GOVERNMENT RESPONSE, INTERNATIONAL SECURITY AND STRATEGIC OUTLOOK

National biological security architectures, international law, European preparedness, economic resilience and the 2027–2031 strategic outlook

Chapter 7. National and Multilateral Biosecurity: The United States, China, Japan, Italy, France, Germany, United Kingdom and European Union

Biological security is becoming a test of state capacity and institutional coordination

The next stage in the evolution of biological security will be determined not only by advances in artificial intelligence and biotechnology but by the capacity of governments to translate scientific information, legal authority and emergency preparedness into functioning systems of prevention, detection and response. The strategic problem is fundamentally institutional. Advanced biological capabilities are distributed across universities, medical laboratories, pharmaceutical manufacturers, commercial technology providers, agricultural research organizations and public-health authorities, while responsibility for preventing deliberate misuse remains divided among ministries, regulators, intelligence agencies, law-enforcement bodies and international institutions. Governments consequently face a challenge that extends beyond conventional counterterrorism: they must preserve the economic and scientific benefits of advanced biotechnology while ensuring that increasingly accessible computational capabilities do not undermine existing arrangements for biological risk management.

The World Health Organization’s framework for deliberate events provides a particularly important starting point because it distinguishes the public-health management of a biological incident from the criminal investigation required to determine whether harmful conduct was intentional. A biological emergency may initially be indistinguishable from a naturally occurring outbreak, and the cause of an event may remain uncertain even after authorities have begun taking protective action. Public-health agencies are responsible for identifying and managing consequences, while security and law-enforcement institutions investigate responsibility and prevent further criminal activity. These functions must operate simultaneously without allowing the requirements of criminal attribution to delay necessary medical intervention. Deliberate Events — World Health Organization, April 2024. Organizzazione Mondiale della Sanità

This division of responsibilities has major implications for artificial intelligence. AI-related biological risk assessments can inform preventive governance, but they do not replace clinical surveillance, epidemiological analysis, laboratory diagnostics, medical stockpiles or emergency management. Equally, an effective health emergency response cannot establish whether the incident was accidental, natural or deliberate without appropriate investigative evidence. A mature national strategy must therefore protect the independence of scientific judgments, establish clear lines of decision-making authority and maintain cooperation among institutions whose responsibilities overlap without becoming interchangeable.

The international environment complicates this requirement. Biological research and pharmaceutical manufacturing operate through transnational supply chains, while legal powers remain primarily national. A government may have strong domestic laboratory regulations but limited authority over research conducted abroad, software supplied through foreign services or critical medical products produced in another jurisdiction. Conversely, international cooperation may improve preparedness without creating binding authority to compel another country to disclose sensitive information or release essential supplies. The resulting architecture must be assessed according to actual legal competence, operational readiness and implementation rather than the number of strategies or agreements announced.

The Biological Weapons Convention and the limits of international enforcement

The Biological Weapons Convention (BWC) provides the central international legal prohibition against biological and toxin weapons. Its obligations were developed around the principle that the hostile purpose of biological activities, rather than the mere existence of biological knowledge, determines the fundamental boundary between prohibited weapons activity and legitimate peaceful research. Article I prohibits the relevant categories of biological agents, toxins, weapons, equipment and delivery systems, while Article III addresses transfer and assistance and Article IV requires States Parties to undertake national measures to prohibit and prevent prohibited activities within their jurisdiction or control.

This distinction is essential because the Convention does not prohibit biotechnology as such. Public-health research, diagnostics, vaccine development, agricultural science and other peaceful applications remain legitimate, subject to applicable national regulations. The Convention therefore establishes a framework for preventing hostile misuse while preserving scientific activity. The United Nations Office for Disarmament Affairs explains that Article IV implementation requires domestic action and coordination among relevant government departments, with different States Parties adopting measures suited to their respective legal systems. The Biological Weapons Convention: An Introduction — United Nations Office for Disarmament Affairs. geneva-s3.unoda.org

The challenge for the 2027–2031 period is that the international prohibition exists alongside major differences in national implementation, transparency, monitoring and enforcement. The Convention is not supported by an operational verification institution equivalent to the Organisation for the Prohibition of Chemical Weapons. Consequently, treaty participation cannot be treated as proof that all relevant biological activities within a jurisdiction are consistently monitored or that potential violations would be detected promptly. Confidence-building measures and diplomatic engagement contribute to transparency, but their existence does not eliminate uncertainties concerning undisclosed programmes or the capacity of individual states to investigate suspected misuse.

The application of the Convention to contemporary artificial intelligence is also not as simple as imposing a general prohibition on biological AI. Computational tools may serve extensive peaceful purposes, and possession of a model with advanced scientific capabilities does not independently establish prohibited biological weapons activity. A defensible legal approach must distinguish the capabilities of general-purpose technology from evidence concerning its intended or actual prohibited use. This is particularly important where governments consider restrictions on software, scientific publication or international research cooperation, because measures extending beyond established legal authority may create unintended consequences for legitimate scientific development.

The principal multilateral requirement is therefore to improve national implementation and international coordination without confusing broad diplomatic commitments with demonstrated enforcement capacity. The Convention establishes the normative foundation, but effective prevention depends on domestic institutions, research governance, appropriate investigative powers and cooperative mechanisms capable of responding to emerging scientific developments.

United States — Research security, national biodefence and federal accountability

The United States possesses an extensive biological security ecosystem that combines federal research funding, public-health surveillance, biotechnology research, medical countermeasure development, defence capabilities and criminal investigation. Its principal strategic advantage is the breadth of these institutions and the scale of the scientific and industrial activity they support. However, this breadth also creates coordination challenges because responsibilities are distributed across agencies with different mandates, authorities, budgets and relationships with private industry.

An important change in the United States’ regulatory direction occurred on 5 May 2025, when the White House issued the executive order Improving the Safety and Security of Biological Research. The order called for stronger oversight of specified categories of federally funded biological research, revisions to existing federal policies concerning dual-use research and research with enhanced pandemic potential, and changes to the framework governing screening in relevant scientific procurement. It also directed consideration of risks arising in research conducted outside federal funding arrangements. The order is significant because it places research security within a broader national policy concerning biological risk, federal accountability and the relationship between scientific innovation and national security. Improving the Safety and Security of Biological Research — White House, May 2025. The White House

The legal and administrative meaning of the order requires precision. A presidential directive can establish executive policy, direct agencies to develop regulations or guidance and influence federal funding conditions, but the issuance of an order does not by itself demonstrate that every subsequent implementation measure has been completed. Its provisions must be distinguished from agency-level rules, funding requirements and independently verified enforcement outcomes. This is particularly important when comparing the United States with jurisdictions that rely on statutory requirements or different institutional mechanisms for research oversight.

The United States also faces an important federalism and market-governance problem. Scientific research occurs across publicly funded universities, private biotechnology companies, hospitals, commercial laboratories and independent research institutions. Federal funding conditions can influence a substantial portion of the research environment, but they are not necessarily equivalent to comprehensive statutory control over every relevant private activity. Effective governance must therefore address the relationship between federal agencies, state authorities, professional institutions and commercial service providers.

The economic dimension is equally significant. Restrictions that are insufficiently precise may affect legitimate biomedical research, increase administrative costs or alter international research relationships. Conversely, insufficient attention to research security can expose public institutions to risks extending beyond an individual project. The policy challenge is to establish requirements that are proportional to demonstrated risks and compatible with continuing scientific innovation. This calls for transparent definitions, effective oversight, appropriate review procedures and the ability to revise measures when evidence changes.

Artificial intelligence adds a further institutional dimension because AI developers and biotechnology organizations do not necessarily operate under the same regulatory frameworks. National security agencies may evaluate advanced AI capabilities, while health authorities govern research practices and public-health agencies prepare for biological emergencies. The principal coordination requirement is to ensure that findings from these different domains can inform appropriate decisions without treating AI evaluation results as direct evidence of biological weapons capability.

China — Biological security as an integrated national security and development responsibility

China’s approach is distinguished by the explicit incorporation of biological security into a comprehensive national legal framework. The Biosecurity Law of the People’s Republic of China, adopted on 17 October 2020 and effective from 15 April 2021, establishes legal provisions covering major infectious disease risks, biotechnology research and development, laboratory biosafety, biological resources, threats involving biological terrorism and biological weapons, and the development of national biosecurity capabilities.

The structure of the legislation is significant because it treats biological security as a cross-sector matter linked to public health, technological development, ecological protection and national security. Chapter VII specifically concerns the prevention of biological terrorism and biological weapons threats, while other chapters govern distinct aspects of scientific research, biosafety and national responsibilities. The law therefore provides an explicit statutory framework for integrating several areas of biological security within a national governance structure. Biosecurity Law of the People’s Republic of China — National Natural Science Foundation of China, official text, April 2021. nsfc.gov.cn

For comparative analysis, however, the existence of a comprehensive statute must not be confused with independently established implementation quality. Legislative scope identifies responsibilities and establishes obligations; it does not by itself demonstrate how consistently those requirements are applied across institutions or how effectively different authorities coordinate during emergencies. Reliable assessment of implementation requires evidence concerning subordinate regulations, inspection outcomes, enforcement activity, institutional capacity and practical preparedness.

China’s scientific and industrial position also creates important questions concerning the interaction between biotechnology development and national security. A governance system seeking simultaneously to promote scientific innovation, protect biological resources and control sensitive activities must balance objectives that can generate institutional tensions. The concentration of regulatory authority may facilitate nationwide policy coordination, but effective oversight still depends on reliable information, professional competence and the quality of enforcement.

From an international perspective, China’s Biosecurity Law provides a basis for comparing statutory obligations with the approaches adopted by the United States and European governments. Such comparisons must be made carefully because the legal definitions, administrative arrangements and available public reporting differ. It would be misleading to infer superior or inferior biological security performance from the mere existence of a broadly worded legal instrument.

The most important issue for the next five years is whether governments can develop sufficiently compatible approaches to evaluating dual-use research, maintaining biological security and supporting international cooperation without undermining legitimate scientific exchange. China’s approach illustrates how extensively biological risk can be incorporated into national legislation, while also demonstrating the need to distinguish formal legal coverage from independently observed operational outcomes.

Japan — Institutional reform, public-health intelligence and the lessons of domestic terrorism

Japan occupies a distinctive position in this assessment because it combines the historical experience of Aum Shinrikyo with substantial institutional reforms intended to strengthen national preparedness for infectious disease emergencies. Its present system cannot be understood solely through the counterterrorism lessons of the 1990s, because the country has also reorganized parts of its public-health and scientific infrastructure in response to subsequent health emergencies.

The most significant recent development is the creation of the Japan Institute for Health Security (JIHS) on 1 April 2025, through the integration of the National Institute of Infectious Diseases and the National Center for Global Health and Medicine. The new institution was established to combine scientific research, infectious disease intelligence, clinical capabilities, crisis response, human-resource development and international cooperation within a more integrated organizational arrangement. The Japanese Ministry of Health, Labour and Welfare identifies information collection, analysis and risk assessment as central functions, alongside research and development, medical care and international engagement. Japan Institute for Health Security — Ministry of Health, Labour and Welfare; About JIHS — Japan Institute for Health Security. 厚生労働省+1

The institutional significance lies in the attempt to connect scientific expertise more directly with government decision-making and clinical response. Historically, public-health research institutions and medical service organizations may possess complementary capabilities but operate through different administrative systems. Their integration creates opportunities for more coherent intelligence and operational support, although the effectiveness of such integration depends on organizational implementation and cannot be inferred from the legal establishment of the new institution alone.

The contrast with Aum Shinrikyo is analytically important but should not be overstated. The historical terrorism case illustrates the dangers of deficient internal expertise and unreliable organizational decision-making, whereas JIHS represents a legitimate public institution designed to strengthen scientific analysis and health emergency management. These are fundamentally different organizational contexts. The relevant lesson is not that one institution directly corrects the other, but that the effective integration of specialized knowledge, professional accountability and crisis decision-making is central to both threat assessment and national resilience.

For Japan, the strategic challenge is to ensure that scientific intelligence, public-health surveillance and the relevant security authorities can cooperate when the origin of a biological event is uncertain. A highly capable epidemiological institution can support the rapid identification and characterization of health threats, while criminal attribution remains a separate legal and investigative responsibility. The country therefore illustrates the importance of maintaining both specialized scientific excellence and institutional mechanisms connecting different branches of government.

The 2026 statements issued by JIHS leadership also provide evidence of continuing institutional consolidation. They describe the first year following the merger as a period of building an integrated operating structure across research, clinical care, epidemiology, public health and international cooperation. These are reported implementation activities, not proof that every intended capacity has reached mature operational performance. Message from the President and Vice President — Japan Institute for Health Security, April 2026. jihs.go.jp

Italy — National planning, regional implementation and the challenge of territorial consistency

Italy’s biological preparedness architecture is shaped by the organization of its National Health Service, the distribution of responsibilities between central government and the regions, and the need to coordinate public-health action across geographically and administratively diverse territories. This creates an important governance distinction between national planning and regional execution. A strategy adopted at the national level may establish common objectives, but its practical effectiveness depends on the ability of individual regions and autonomous provinces to translate those objectives into functioning surveillance, prevention, clinical and emergency arrangements.

Two major developments in 2026 are relevant. On 30 April 2026, the Italian Ministry of Health announced approval of the National Pandemic Plan 2025–2029, describing it as a five-year framework intended to support preparedness and response to respiratory pandemics with different epidemiological characteristics. The ministry emphasized that the plan is not limited to a single pathogen and is intended to support a more general response methodology. This is an important step in pandemic preparedness, although such a plan should not be described as a comprehensive national counter-bioterrorism strategy simply because some health-system functions may also be relevant to deliberate incidents. Approval of the National Pandemic Plan 2025–2029 — Italian Ministry of Health, April 2026. Ministero della Salute

A second development occurred on 21 May 2026, when the State–Regions Conference adopted the National Prevention Plan 2026–2031. The Ministry of Health described the plan as the principal framework for prevention and health promotion across the national territory, with regional prevention plans required to implement the agreed objectives. The ministry also announced an additional €50 million for 2026 to support territorial prevention activities and associated implementation requirements. This figure is a specific addition to prevention resources, not a dedicated biological terrorism allocation, a five-year programme total or evidence that the amount has already been fully disbursed. National Prevention Plan 2026–2031 — Italian Ministry of Health, May 2026. Piano Nazionale della Prevenzione 2026-2031

The significance for biological security is primarily institutional. Italy’s national planning arrangements offer mechanisms through which surveillance, prevention and health protection can be coordinated, but the effectiveness of those mechanisms depends on regional capacity and consistent implementation. The presence of legally established objectives does not demonstrate equivalent operational readiness in every territory. Variations in health-system capacity, personnel availability, laboratory arrangements and emergency logistics must therefore be examined through compatible official indicators rather than assumed from the national adoption of a plan.

The country’s economic exposure also requires attention. Italy’s pharmaceutical and biomedical activities, its regional healthcare infrastructure and its participation in European supply chains make biological preparedness relevant to both public health and industrial continuity. A major health emergency could affect the workforce, hospital capacity, procurement requirements and the availability of essential medical products. However, the scale and distribution of these effects cannot be estimated reliably without a defined incident scenario and appropriate economic data. It would therefore be inappropriate to attach an unsupported monetary loss estimate to the threat of AI-enabled biological terrorism.

The interaction with European institutions is particularly important. Italy participates in EU-level arrangements concerning cross-border health threats and medical countermeasure preparedness, while retaining national and regional responsibilities for healthcare delivery. European coordination may improve access to information and resources, but it does not eliminate the need for territorial planning and functioning emergency procedures. The practical test is whether national, regional and European institutions can exchange relevant information and act coherently when a health emergency develops under conditions of uncertainty.

For the 2027–2031 period, the central Italian priority is not the creation of an additional declaratory strategy without evidence of implementation. It is the demonstrable integration of the new national plans into regional readiness, supported by appropriate exercises, reliable information flows and defined institutional responsibilities. The strongest indicators will concern whether the regional arrangements required by the 2026 prevention plan are adopted, assessed and maintained over time, and whether pandemic preparedness procedures can coordinate effectively with security and civil-protection institutions when an incident is suspected to be deliberate.

France — Interministerial NRBC governance and the integration of civilian security with public health

France’s biological security architecture is distinguished by an established national framework for nuclear, radiological, biological and chemical risks, linked to broader national security planning and the coordination of specialized civilian and military capabilities. The Secrétariat général de la défense et de la sécurité nationale (SGDSN) plays an important role in coordinating interministerial arrangements, while the public-health system maintains its own responsibilities for prevention, clinical care and emergency response. This structure recognizes that a biological incident may simultaneously require scientific expertise, emergency medical care, public-order measures and national crisis coordination.

The SGDSN describes a national NRBC-E arrangement involving strategic coordination, preventive regulation and intervention capabilities. Its public documentation identifies a strategic committee chaired by the SGDSN, alongside operational responsibilities involving police, gendarmerie, firefighters, emergency medical services and specialized national resources. It also identifies a national civilian and military training centre intended to strengthen preparation and interoperability among relevant services. These institutional elements provide evidence of a structured national approach, although their existence does not automatically establish the level of readiness achievable in a particular emergency. Improving Protection and Prevention Arrangements — SGDSN, France. SGDSN

France’s health emergency arrangements operate through the ORSAN system, which organizes the response of healthcare institutions to exceptional health situations. The framework includes five operational plans addressing different categories of emergency, including the ORSAN REB plan for epidemic and biological risks. Its purpose is to coordinate the mobilization of health facilities and relevant healthcare personnel, including the ability to expand clinical and diagnostic capacity when required. The arrangements are developed through regional health agencies and must interact with broader civil-protection mechanisms. The ORSAN Framework for Exceptional Health Situations — French Ministry of Health. sante.gouv.fr

The significance of this dual structure is that the health-system response to a biological emergency is not treated as identical to the national security response, even though the two may need to operate simultaneously. A suspected deliberate incident can require immediate clinical and epidemiological intervention while investigative authorities assess possible criminal responsibility. The institutional challenge is to ensure that these parallel processes are sufficiently coordinated without compromising the evidentiary independence or legal responsibilities of either.

France’s arrangements also have an industrial and technological dimension. Advanced medical research, pharmaceutical production, diagnostics and specialized scientific expertise contribute to national resilience, while also requiring appropriate governance of dual-use technologies. The country’s strategic interest therefore includes both the protection of public health and the preservation of scientific and industrial capabilities. Measures concerning advanced AI and biotechnology should be assessed for their consequences across these different objectives rather than treated exclusively as restrictions on technical activity.

For France, the most relevant evidence of readiness will be the demonstrated performance of interministerial exercises, the interoperability of emergency services, the ability of regional health institutions to expand capacity and the effectiveness of coordination between scientific advisers and crisis authorities. These criteria offer a more useful basis for assessment than the assumption that centralized national planning alone guarantees consistent performance across every region.

Germany — Federal public-health responsibilities and specialized biological threat expertise

Germany’s preparedness system combines federal scientific institutions with important responsibilities exercised by the Länder and local authorities. This arrangement creates a different coordination environment from highly centralized national systems because the identification and management of health incidents may involve authorities operating at several governmental levels. The central challenge is to ensure that specialized scientific expertise is available to those responsible for practical decisions while preserving the legal and administrative responsibilities of federal, state and local institutions.

The Robert Koch Institute (RKI) provides a key scientific component through its Centre for Biological Threats and Special Pathogens (ZBS). The RKI describes the ZBS as responsible for identifying and assessing biological threat situations, including incidents involving highly pathogenic or bioterrorism-relevant agents, and for developing approaches to the prevention and management of their public-health consequences. Its responsibilities include scientific assessment, expert information and support to decision-makers. Importantly, the centre’s work extends beyond deliberate incidents to natural and accidental outbreaks involving relevant biological hazards. Centre for Biological Threats and Special Pathogens — Robert Koch Institute, updated August 2026. rki.de

The RKI also describes arrangements for specialist assistance during biological incidents, including advice to emergency services and public-health authorities. These arrangements demonstrate the importance of scientific support to operational responders without implying that every incident will require the same resources or follow an identical response sequence. Robert Koch Institute as a Partner Authority in the CBRN Support Network — RKI. rki.de

Germany’s institutional structure makes interoperability a particularly important consideration. An incident may first be identified through local healthcare or public-health channels, while the scientific expertise necessary for detailed assessment is concentrated in specialized national institutions. Coordination therefore depends on clear reporting arrangements, professional relationships, established procedures and the ability to obtain relevant technical support within the appropriate timeframe.

The economic dimension is also substantial. Germany’s pharmaceutical, chemical, medical-technology and research sectors are important components of its industrial economy and of European supply chains. Biological preparedness must therefore account for both the possible disruption caused by an emergency and the consequences of regulations affecting legitimate research and production. However, these considerations do not justify presenting the scale of the country’s industrial sector as a direct measure of biological security readiness. Industrial capability, emergency availability and scientific expertise are related but distinct categories.

For the 2027–2031 horizon, the most consequential issue is the operational interface between scientific specialization and distributed governmental responsibilities. Germany’s institutional architecture contains clearly identifiable technical capabilities, but the quality of preparedness must be assessed through exercises, implementation evidence, consistent reporting and the ability of federal and regional institutions to coordinate under pressure.

United Kingdom — An integrated biological security strategy with measurable implementation reporting

The United Kingdom provides a particularly useful case because it has combined a published national biological security strategy with successive implementation reports and explicit commitments extending to 2030. The UK Biological Security Strategy, published in June 2023, organizes national activity around four functions: understanding biological risks, preventing them where possible, detecting and characterizing emerging threats, and responding to reduce their consequences. The framework also identifies cross-cutting responsibilities concerning national leadership, scientific expertise, industrial capabilities and international cooperation. UK Biological Security Strategy — Cabinet Office, June 2023. GOV.UK

The importance of this structure lies not merely in its conceptual coherence but in the availability of more recent implementation evidence. On 14 July 2026, the Cabinet Office published its Biological Security Strategy Implementation Report covering July 2025 to July 2026. The report provides information on achievements, outstanding commitments and the government’s approach to biological surveillance. This creates a more developed public record than a strategy document alone, although the reported achievements remain government statements and require independent assessment where operational effectiveness is material. UK Biological Security Strategy: Implementation Report, July 2025–July 2026 — Cabinet Office, July 2026. GOV.UK

A parliamentary statement published on the same date reported several specific developments. The government identified £1.83 billion of investment supporting a network of National Biosecurity Centres, described the operation of a Biothreats Radar intended to improve situational awareness, and reported that the new pandemic preparedness strategy was supported by an additional £1 billion in health-protection measures. These figures refer to separate programmes or investment descriptions and must not be added together without establishing their precise fiscal coverage, accounting definitions and potential overlaps. They should not be interpreted as expenditure solely dedicated to preventing AI-enabled biological terrorism. UK Government Resilience and Biological Security Strategy Implementation Reports — UK Parliament, Written Statement HCWS244, July 2026. UK Parliament

The Biothreats Radar is significant because the government describes it as an operational capability for integrating biological threat information across public institutions. The implementation report states that the system was being used by 14 government departments and agencies and had been exercised during national preparedness activities. This is evidence of deployment and reported institutional use, rather than independent proof that the system can detect every relevant threat or prevent deliberate biological incidents. The distinction between deployment, operational utilization and demonstrated detection effectiveness remains essential. UK Biological Security Strategy Implementation Report, 2025–2026 — Cabinet Office, July 2026. GOV.UK

The United Kingdom’s experience illustrates the advantages of reporting implementation progress in a manner that makes administrative commitments more visible. It also demonstrates the continuing challenge of evaluating real-world effectiveness. A programme may be operational and widely used without a public record sufficient to determine its performance across all relevant scenarios. The appropriate indicators consequently include interoperability, independent exercise evaluation, reporting completeness and the ability to connect scientific analysis with national decision-making.

European Union — Cross-border health security, industrial preparedness and shared competence

The European Union’s biological preparedness architecture is particularly important because it connects the national health systems of Member States with EU-level responsibilities concerning cross-border threats, scientific coordination and medical countermeasure preparedness. The division of competence means that European institutions cannot be treated as possessing identical authority to national governments. Member States retain essential responsibilities for healthcare organization and delivery, while EU mechanisms support coordination, common assessment and preparedness activities within their respective legal mandates.

The Health Emergency Preparedness and Response Authority (HERA) is a central element of this framework. HERA supports the assessment of threats, identification of medical countermeasure requirements, monitoring of supply chains and development of industrial readiness. Its functions operate differently during preparedness and crisis phases, reflecting the distinction between long-term planning and the activation of emergency measures under the applicable EU framework. HERA Overview — European Commission. Public Health

The Commission’s published account of threat assessment describes the use of open-source epidemic intelligence, improvements in laboratory and diagnostic capabilities, and ongoing efforts concerning information systems for health-threat assessment. It also reports that HERA’s prioritization processes have identified more than 500 medical countermeasure items relevant to the assessed threat categories. This figure represents a defined planning and prioritization list, not 500 products already purchased or immediately available in operational stockpiles. The distinction between a requirement, an approved procurement, an available stock and a deployable medical product is essential for evaluating genuine preparedness. Threat Assessment and Intelligence Gathering — European Commission, HERA. Public Health

The Commission’s current HERA programme documentation also identifies approximately €1.7 billion of Horizon Europe funding earmarked for HERA within the wider programme for 2021–2027. This allocation must be interpreted according to the relevant programme definitions and should not be treated as a dedicated EU biological counterterrorism budget. It reflects support for a wider medical countermeasure and preparedness agenda. Funding and Opportunities — European Commission, HERA. European Commission

The strategic importance of HERA is therefore partly industrial. Biological emergencies can generate simultaneous demand across several Member States, while the production of relevant products depends on manufacturing capacity, raw materials, specialized personnel, regulatory procedures and distribution networks. A common procurement or preparedness mechanism may reduce some coordination problems, but it cannot automatically eliminate industrial bottlenecks or guarantee timely delivery. These limitations require distinction between announced production capacity, contracted supplies and products actually available to health systems.

The EU framework also has implications for artificial intelligence governance, although biological security and AI regulation must not be treated as identical policy domains. AI governance may address system safety, accountability or prohibited uses, while biological security also requires institutions responsible for laboratories, medical products, infectious disease surveillance and emergency response. The practical objective is to ensure appropriate cooperation across these domains without assuming that compliance with one regulatory instrument establishes comprehensive protection against biological misuse.

International coordination and the limits of interoperability

The comparison of national approaches reveals that governments are developing biological preparedness through different combinations of legal instruments, specialized scientific institutions, public-health structures and national security arrangements. China emphasizes a comprehensive national biosecurity statute. Japan has integrated major infectious disease research and clinical institutions. Italy is implementing new national prevention and pandemic planning arrangements through its regional health system. France relies on interministerial NRBC coordination alongside organized health emergency planning. Germany maintains specialized scientific threat expertise within a federal institutional environment. The United Kingdom has developed a national biological security strategy supported by successive implementation reports, while the United States combines extensive scientific and security institutions with changing federal research oversight. The European Union provides coordinating and industrial preparedness mechanisms within a system of shared and divided competence.

These differences should not be converted into a simple ranking of national preparedness. The available records do not use a common measurement system capable of establishing directly comparable performance across the full range of biological risks. A comprehensive statute, a specialized scientific centre, an implementation report and a national exercise are different kinds of evidence. They can establish particular institutional characteristics, but they cannot independently determine which country would perform more effectively in a specific biological emergency.

The WHO’s approach to deliberate events provides an appropriate common standard for analyzing the practical requirements of coordination. Its framework emphasizes that suspected deliberate incidents involve medical, epidemiological, criminal and forensic responsibilities, potentially across multiple jurisdictions. These functions create a requirement for institutional cooperation that cannot be replaced by a single national policy document. Deliberate Events — WHO, April 2024. Organizzazione Mondiale della Sanità

The most consequential international weaknesses are likely to arise where information cannot be exchanged in time, legal responsibilities are unclear, medical products are unavailable despite formal preparedness plans, or institutions are unable to distinguish an uncertain public-health event from a suspected criminal incident without disrupting the necessary response. These are conditional analytical vulnerabilities, not findings that any named country currently lacks a particular classified capability.

Chapter 7 — Key judgments

The countries examined possess materially different legal and institutional approaches to biological security. Their official frameworks establish substantial preparedness activity, but the public record does not support a complete comparative ranking of national effectiveness. The most valuable distinction is between established legal authority, announced institutional capability, reported implementation and independently demonstrated operational performance.

For Europe, the decisive issue is how national arrangements in Italy, France and Germany interact with EU coordination and industrial preparedness, while the United Kingdom maintains separate but potentially complementary arrangements. For the wider international system, the central requirement is improved compatibility between scientific assessment, public-health response and national security responsibilities without eroding legitimate research or public accountability.

What would change the assessment: Comparable official evidence from national exercises, independently assessed preparedness outcomes, revised legal instruments and documented improvements or failures in cross-border coordination.

Open official record: The most important gaps concern comparable operational-readiness indicators, consistent reporting of medical countermeasure availability, detailed information on interagency interoperability and transparent implementation assessments.

Chapter 8. Strategic Scenarios for 2027–2031, Early-Warning Indicators, Policy Options and Final Net Assessment

Strategic forecasting under conditions of incomplete evidence

The principal difficulty in forecasting AI-related biological security risks is that several underlying variables are changing simultaneously, while the available evidence does not permit their effects to be isolated with sufficient precision. Scientific AI capabilities are advancing, biological research tools continue to evolve, governments are updating their regulatory frameworks, and public and private institutions are developing new preparedness arrangements. Yet the relationship between these changes and the practical capabilities of violent non-state actors remains uncertain. A credible strategic forecast must therefore distinguish developments that have been measured from assumptions concerning how those developments might influence future threats.

The 2027–2031 horizon is sufficiently long for material changes in technology, regulation and institutional readiness, but not sufficiently well characterized to justify invented numerical probabilities. Historical incident counts provide a limited basis for estimating future events because they reflect different technologies, different reporting practices and an incomplete observation of relevant activity. Scientific benchmark results likewise cannot be translated directly into predictions about attack frequency. The principal purpose of forecasting is consequently to identify alternative conditions under which biological security may improve, deteriorate or become more difficult to evaluate, and to determine what observable evidence would support a revision of present judgments.

The World Health Organization’s deliberate-event framework reinforces the need to address uncertainty before criminal attribution has been established. Its guidance recognizes that harmful events may be initially ambiguous and that health protection must proceed alongside investigative activity. This creates an important strategic principle: a national preparedness system should be effective across different possible causes of a biological emergency, rather than depend on advance knowledge of the perpetrator or the exact technological means involved. Deliberate Events — World Health Organization, April 2024. Organizzazione Mondiale della Sanità

The following scenarios are therefore conditional planning pathways, not predictions of specific attacks. They are not assigned numerical probabilities because a sufficiently defensible empirical basis is unavailable.

Scenario A — Technological progress without corresponding operational transformation

Under the first pathway, frontier AI continues improving on scientific evaluations and legitimate research applications, but practical biological constraints remain substantial enough to prevent a general transformation of malicious capabilities. Biological research still requires qualified personnel, physical resources, reliable validation and the capacity to interpret uncertain results. Advances in computation therefore produce meaningful benefits within established scientific institutions without eliminating the differences between access to information and the ability to conduct complex physical activities.

This scenario would be consistent with continued progress in AI benchmarks alongside limited evidence that such improvements translate into reliable performance by inexperienced or poorly organized users. It would also be consistent with successful institutional efforts to maintain safeguards, improve scientific oversight and reduce misuse opportunities without broadly restricting legitimate research.

For governments, the principal risk under this pathway would be misallocation of resources through exaggerated assumptions about AI-enabled threats. Excessive emphasis on speculative technological capabilities could divert attention from established public-health vulnerabilities, emergency logistics and preventable infectious disease risks. The appropriate response would involve sustained evaluation, proportionate safeguards and continued investment in capabilities that provide benefits across natural, accidental and deliberate biological emergencies.

The observation that would strengthen this pathway is not simply the absence of a successful attack. It would be reproducible evidence that measured improvements in AI do not substantially alter relevant practical outcomes under appropriately controlled conditions, together with evidence that protective arrangements remain effective. Absence of an observed incident, by itself, would remain insufficient because the underlying frequency of attempts is not reliably known.

Scenario B — Selective improvement in AI-assisted scientific and organizational performance

The second pathway involves selective changes in the performance of users and institutions that already possess relevant scientific or organizational competence. Advanced AI systems may become increasingly useful in legitimate research, scientific interpretation and institutional decision support, while their practical effects remain uneven across different tasks and user populations. The resulting security environment would not necessarily involve the disappearance of conventional barriers. Instead, some activities could become more efficient while others continue to require capabilities that AI cannot independently provide.

This pathway is important because it avoids the false choice between complete technological transformation and no meaningful change. Scientific and organizational capabilities are multidimensional, and improvements need not occur uniformly. A system may become substantially more useful for certain legitimate tasks without becoming reliable in every practical context. Security authorities would therefore need to update assessments according to independently observed improvements rather than relying on a single generalized capability label.

The principal governance challenge would be ensuring that risk evaluations remain sufficiently current to reflect changing model capabilities and deployment arrangements. Centrally hosted systems may permit ongoing intervention by providers, while independently deployed systems present different limitations. Scientific institutions would need to maintain appropriate oversight without assuming that the availability of computational assistance eliminates professional responsibilities.

This pathway would be supported by reproducible evidence of improved legitimate research outcomes in selected domains, alongside persistent limitations in other tasks. The critical distinction would remain between observed assistance and hypothetical malicious consequences. Improvements in legitimate performance would warrant evaluation of potential security implications, but would not independently establish that a terrorist organization possesses a particular capability.

Scenario C — Fragmented governance and uneven institutional preparedness

A third pathway concerns the possibility that technological adoption proceeds more quickly than governments and international institutions can coordinate appropriate safeguards. Under this scenario, different jurisdictions and commercial providers develop divergent approaches to AI evaluation, research oversight and model release. Some institutions maintain strong controls while others lack consistent procedures or sufficient resources to implement them effectively. The principal challenge would be not a single technological breakthrough but the cumulative consequences of uneven governance across an interconnected scientific ecosystem.

The plausibility of this pathway follows from the existing distribution of legal authority and institutional responsibility. The United States, China, Japan, European Union and major European countries employ different legal instruments and administrative structures. Commercial AI providers also differ in their deployment arrangements and their ability to influence downstream use. International agreements establish important principles, but do not eliminate these differences.

The resulting vulnerabilities could include inconsistent research oversight, unclear responsibility for assessing integrated AI systems, insufficiently coordinated emergency preparedness and limited comparability of national reporting. These are governance concerns rather than evidence of a specific criminal capability. The appropriate governmental response would be to improve institutional clarity, develop compatible evaluation terminology, maintain appropriate international cooperation and ensure that relevant safeguards are demonstrably implemented rather than merely announced.

This pathway would gain support from documented failures of institutional coordination, incompatible evaluation standards or evidence that formal preparedness commitments have not resulted in operational capabilities. It would be weakened by transparent implementation records, successful cross-border exercises and demonstrably effective cooperation among relevant institutions.

Scenario D — Stronger integrated preparedness and greater institutional resilience

A fourth pathway involves the progressive integration of biological security into national and international preparedness systems. Under this scenario, scientific AI evaluations become more rigorous, appropriate safeguards are updated as models evolve, public-health authorities improve their detection and response capabilities, and governments develop more effective arrangements for sharing relevant information under lawful conditions. The resulting system would not eliminate biological threats, but could reduce their consequences and improve the reliability of preventive decisions.

Several existing institutional developments provide a foundation for this pathway. The United Kingdom has established a biological security strategy with implementation reporting and operational surveillance initiatives. Japan has created an integrated health security institution. The European Commission is developing medical countermeasure and threat-intelligence capabilities through HERA. Italy has adopted new national prevention and pandemic planning frameworks, while France and Germany maintain specialized structures for biological emergency management. These developments demonstrate institutional investment and reform, although they do not prove that the full resilience envisioned has already been achieved.

The pathway would become more credible if independent exercises demonstrated improved coordination, if medical countermeasure planning translated into verified availability, and if AI safeguards were shown to remain robust under appropriately controlled testing. It would also depend on maintaining sufficient investment, scientific expertise and political attention beyond the immediate period following major health emergencies.

The principal economic benefit would arise from reducing uncertainty and strengthening continuity across health systems, research institutions and essential supply chains. However, the precise financial value of that resilience would depend on the threats considered, the counterfactual assumptions adopted and the measured effectiveness of the relevant interventions. A defensible financial estimate would require more complete data than are presently established in this assessment.

Early-warning indicators must measure observable change rather than assumed danger

An effective early-warning framework should focus on developments that can be observed, independently assessed and connected to a defined policy decision. It should not depend on speculative numerical threat scores or interpret the ordinary use of artificial intelligence as evidence of malicious intent. The central objective is to recognize when assumptions about technology, governance or preparedness are no longer supported by the evidence.

The first category concerns scientific capability evaluation. Relevant indicators include reproducible changes in the performance of frontier AI systems on legitimate scientific tasks, the reliability of model-generated assistance and the extent to which results are confirmed outside the original evaluation setting. These indicators should be interpreted according to their actual methods and should not be converted into direct measures of terrorist capability.

The second category concerns safeguard robustness. Developers and qualified evaluators should examine whether protective systems remain effective across relevant deployment arrangements, including whether documented weaknesses are addressed and whether changes in model capabilities require revised controls. The existence of a vulnerability should be distinguished from evidence of real-world exploitation or harm.

The third category concerns institutional preparedness. Governments can monitor the completion of planned exercises, the implementation of national strategies, the availability of relevant medical products, the effectiveness of interagency communication and the capacity of public-health institutions to manage uncertain biological events. These measures provide direct information about resilience, although they must be supported by transparent definitions and independent review where practicable.

The fourth category concerns international coordination. Observable developments include changes in national legal implementation, the adoption of compatible evaluation frameworks, the functioning of information-sharing arrangements and the completion of cross-border preparedness activities. These indicators are particularly relevant to the European Union and its neighboring partners because biological emergencies and essential supply chains do not necessarily respect national boundaries.

The fifth category concerns economic and industrial continuity. Medical countermeasure availability, manufacturing readiness, procurement execution and logistical resilience should be monitored separately from announced financial commitments. An allocated budget is not equivalent to money disbursed, and contracted production capacity is not equivalent to deployable stock. Governments require evidence concerning actual availability and time-sensitive response capability before concluding that an industrial preparedness objective has been achieved.

Economic resilience and the cost of biological uncertainty

Biological security has consequences extending beyond direct medical expenditure. A significant biological emergency can disrupt labor availability, public services, transport, education, commercial activity and international trade. The magnitude of these effects depends on the nature of the incident, the measures required to manage it, public behavior, institutional communication and the capacity of the healthcare system. An intentional event may also create investigative and security demands that would not arise to the same degree in an otherwise comparable natural outbreak.

These consequences make preparedness a legitimate economic policy concern, but they do not justify transferring financial estimates from historical pandemics directly to hypothetical acts of terrorism. The scale, duration and geographic distribution of consequences may differ substantially. A credible economic assessment should identify the relevant scenario, distinguish direct from indirect effects, specify the expected duration of disruption and establish which costs would be reduced by particular preparedness measures.

The industrial dimension also requires attention to how investments are classified. Funding for public-health prevention, pandemic preparedness, biotechnology research, emergency logistics and national security may contribute to related objectives without being interchangeable. The additional €50 million identified in Italy’s 2026 prevention planning, the United Kingdom’s reported investments in biosecurity and health protection, and the European Union’s medical countermeasure funding illustrate materially different fiscal instruments and institutional purposes. They cannot be aggregated into a single international biological security spending total without resolving their definitions, time periods and possible overlaps. Italian National Prevention Plan 2026–2031 — Ministry of Health; UK Biological Security Strategy Implementation Statement — UK Parliament, July 2026; HERA Funding and Opportunities — European Commission. Piano Nazionale della Prevenzione 2026-2031+2

For financial authorities, the principal objective should be to distinguish expenditures that improve general health-system resilience from those addressing specific security vulnerabilities. Measures that provide benefits across natural, accidental and deliberate incidents may offer broader economic value than narrowly defined interventions, but their comparative cost-effectiveness must be demonstrated through appropriate analysis rather than asserted.

Policy choices: authority, feasibility and unintended consequences

The first policy option is to strengthen evidence-based evaluation and reporting. Governments, scientific institutions and qualified independent evaluators can establish common terminology for model capability, experimental reliability and safeguard effectiveness. The expected benefit is improved comparability and fewer unsupported conclusions. Implementation requires specialized personnel, access arrangements and appropriate protection of sensitive findings. The principal risk is that poorly designed reporting requirements may impose excessive burdens or encourage the disclosure of information that should be handled responsibly.

A second option is to improve national biological preparedness through cross-agency exercises and institutional accountability. This can be pursued through existing health, security and civil-protection authorities, depending on the jurisdiction. The expected benefit is improved understanding of operational responsibilities and earlier identification of coordination problems. Such activities require recurrent funding and staff commitment, and their usefulness depends on whether findings are translated into corrective measures. Exercises that are primarily ceremonial provide limited evidence of resilience.

A third option concerns responsible AI deployment and research governance. Governments and developers can adopt safeguards proportionate to verified capabilities and deployment risks, while scientific institutions maintain professional oversight of dual-use research. The expected effect is to reduce misuse opportunities without unnecessarily restricting legitimate innovation. However, excessive or poorly targeted restrictions may impair scientific competitiveness, prevent beneficial research or encourage activity to move into less transparent environments. The relevant policy test is therefore proportionality supported by evidence.

A fourth option is to strengthen medical countermeasure and supply-chain preparedness. National authorities and relevant European institutions can improve the quality of procurement planning, verify the availability of essential products and assess whether manufacturing arrangements are operationally credible. The expected benefit is improved capacity to manage emergencies independently of their cause. The implementation burden includes procurement expenditure, storage requirements, industrial coordination and possible obsolescence of stockpiled products. Long-term preparedness therefore requires regular review of requirements and actual availability.

A fifth option is to improve international cooperation within defined legal mandates. States can exchange appropriate scientific and public-health information, support treaty implementation and develop compatible emergency coordination procedures. Such arrangements may reduce delays and uncertainty during cross-border events, but their effectiveness depends on trust, legal compatibility and the protection of sensitive information. International mechanisms should complement national responsibilities rather than create uncertainty about which authority is empowered to act.

The strongest policy architecture is not necessarily the most restrictive. It is the one capable of demonstrating that its interventions reduce relevant risks, remain legally supportable and impose proportionate costs on legitimate scientific and economic activity.

Final cross-pillar assessment

The three pillars of this dossier establish a progression from historical evidence to technological uncertainty and finally to the responsibilities of government. The historical record demonstrates that violent intention, scientific resources and organizational ambition do not automatically produce effective biological capabilities. Contemporary AI evaluations demonstrate advances in scientific knowledge processing and selected forms of research assistance, while leaving unresolved questions about practical operational effects. National and international institutions have developed substantial legal and preparedness arrangements, but their effectiveness varies according to implementation, information quality, operational coordination and the ability to adapt when scientific and technological conditions change.

The central strategic error would be to interpret any one of these observations in isolation. Historical failure does not establish permanent technological immunity. Scientific progress does not establish that a violent organization has acquired an effective biological weapon. A national strategy does not prove operational readiness, and a formal prohibition does not guarantee enforcement. Conversely, uncertainty concerning future misuse does not justify treating legitimate scientific research as inherently dangerous or assuming that every new AI capability requires the same form of restriction.

The appropriate government response is therefore based on differentiated evidence and layered institutional responsibility. AI developers should be assessed according to model capabilities and the effectiveness of relevant safeguards. Scientific institutions should maintain responsible research governance. Public-health authorities should preserve the capacity to identify and manage biological incidents regardless of cause. Law-enforcement and security agencies should investigate credible threats within applicable legal frameworks. Governments and international institutions should ensure that these functions remain coordinated without confusing their different responsibilities.

The decisive strategic judgment for 2027–2031 is that biological security will increasingly depend on the ability of states to manage uncertainty across scientific, technological, institutional and economic systems. Governments that can distinguish measured capabilities from unsupported claims, integrate legitimate scientific expertise into decision-making and demonstrate practical preparedness will be better positioned to respond to changing biological risks. The objective is not to predict a particular future attack, but to ensure that public institutions can recognize consequential changes, apply proportionate preventive measures and respond effectively when biological emergencies occur.

Chapter 8 — Key judgments

The present evidence supports a qualitative forecast based on conditional scenarios, not a numerical prediction of successful biological terrorism. AI capability growth, deployment arrangements, research governance and institutional preparedness may evolve in different directions. Governments should therefore monitor observable changes and revise assessments when credible evidence warrants doing so.

The most consequential policy options involve independent evaluation, proportionate safeguards, clear institutional accountability, practical preparedness exercises, reliable medical countermeasure availability and international cooperation. Their value should be assessed through implementation evidence and measurable resilience rather than the existence of plans or public announcements.

What would change the final assessment: Independently corroborated evidence of material changes in practical scientific capabilities, substantial new international legal commitments, reproducible improvements in safeguard effectiveness or documented changes in national preparedness that alter the balance between risk exposure and institutional response capacity.

Final net assessment: The history of biological terrorism remains indispensable to understanding the relationship between intent and capability, but it cannot function as a permanent forecast. The convergence of artificial intelligence and biotechnology makes institutional adaptability, scientific integrity and operational preparedness increasingly important elements of national and international security.

International Biological Security Architecture, National Comparison and 2027–2031 Outlook

The institutional evidence is based on the Biological Weapons Convention — UNODA, WHO guidance on deliberate biological events, UK Biological Security Strategy, European Commission HERA, US federal biological research policy, China’s Biosecurity Law, Japan Institute for Health Security, Italian Ministry of Health, French SGDSN and Germany’s Robert Koch Institute.

PILLAR III · GOVERNMENT AND INTERNATIONAL SECURITY

International Biological Security Architecture

National institutions, multilateral coordination, four strategic pathways and preparedness indicators.

JURISDICTIONS EXAMINED
8
Seven national frameworks and the European Union.
STRATEGIC HORIZON
2027–31
Five-year assessment period, not an attack forecast.
QUANTIFIED ATTACK PROBABILITY
N/A
No evidence-supported numerical probability.

Comparative National and European Architecture

Select a jurisdiction to examine its official institutional framework, governance characteristics and evidence limitations. The comparison does not rank national preparedness.

INSTITUTIONAL PROFILE

Official framework

Security and resilience implication

Evidence limitation

Jurisdiction Principal framework Governance structure Evidence qualification
Source governance: BWC/UNODA, WHO, national government strategies and institutions, European Commission HERA. Historical frameworks and current institutional positions are distinguished from future scenarios. No live data, readiness rankings or attack probabilities are generated.

The comparison deliberately avoids national rankings or fictitious preparedness measurements. The scenario view identifies four possible, overlapping developments rather than assigning unsupported probabilities, while the monitoring table specifies the official evidence needed to assess actual progress.


Copyright of debuglies.com – Even partial reproduction of the contents is not permitted without prior authorization – Reproduction reserved

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Questo sito utilizza Akismet per ridurre lo spam. Scopri come vengono elaborati i dati derivati dai commenti.