Scope. This dossier examines NATO, the European Union, Italy, France, Germany and the United Kingdom from 2024 to 18 November 2026, assessing how competition among advanced AI models could change military cyber defence, cyberterrorism and cognitive warfare through November 2031.

Executive Summary / BLUF

  • AI’s clearest near-term military value is faster, verified cyber defence. It can accelerate software review and incident analysis. The operational advantage depends on whether teams can validate findings, close exposures and restore services faster under pressure.
  • OpenAI reported on 17 November 2026 that NATO had used up to 300 billion raw input and output tokens per month under its work with the NATO Communications and Information Agency. The provider’s figure measures usage. The public announcement provides no corresponding measure of security outcomes, cost or sustained operational readiness. Strengthening NATO and allied cyber defenses with AI — OpenAI — Nov 2026. OpenAI
  • Crossed Swords 2026 brought together more than 250 participants from over 40 NATO and non-NATO countries in Tallinn on 2–6 November. The exercise used both on-premises AI and Daybreak and introduced an AI-agent training stream with human oversight. It establishes exercise use. Operational effectiveness remains unquantified in the organiser’s account. Exercise Crossed Swords tests cyber operations and multi-domain coordination — NATO CCDCOE — Nov 2026. ccdcoe.org
  • The threat is uneven. The strongest documented terrorism-related use concerns propaganda and hate speech. In foreign information manipulation, the EEAS detected AI-related activity in 147 of 540 incidents in 2025, while reporting low organic engagement for much of the synthetic material it observed. EU TE-SAT 2025 Executive Summary — Europol — Jun 2025. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — Mar 2026. europol.europa.eu
  • The procurement decision is whether gains survive operational constraints. Wider deployment should depend on measured improvements in remediation, recovery and decision quality under degraded conditions, with auditable human authority and a tested fallback.

NATO’s AI Advantage Will Be Decided After the Model Answers

On 17 November 2026, OpenAI disclosed the scale of NATO’s use of its models for cyber defence through the NATO Communications and Information Agency. The same day, EU defence ministers endorsed a secure European defence cloud as a candidate for joint development. The two decisions expose a choice that defence budgets will have to resolve. The Alliance can acquire powerful models quickly, but its military advantage depends on whether national authorities can turn their findings into verified action when a network is degraded, a supplier is unavailable or an ally needs evidence it is authorised to use. That requirement puts procurement, industrial capacity and diplomatic trust on the same balance sheet. Buying model access is the simplest transaction. Paying for the ability to act on its output is the harder commitment.

The figures show where the money must go

OpenAI reported that NATO had used up to 300 billion raw input and output tokens per month in the work with the NATO Communications and Information Agency. The November disclosure establishes a provider-reported peak. It supplies no corresponding figure for expenditure, verified fixes or time saved. The United Kingdom’s October 2026 announcement comes closer to an operational measure. The government said two pilots had helped identify and fix more than 1,000 potential vulnerabilities across government systems. Its published account describes departments validating and prioritising the findings through existing procedures. It leaves open how the work compared with ordinary practice by severity, review time and time to closure. Poland provides a different form of evidence. In September 2026, CERT Polska identified the models used in research behind a published security advisory and directed operators to available vendor fixes. That traceable chain from research to remediation is a better basis for procurement than the volume of computation alone.

The infrastructure commitments show how quickly the surrounding costs can exceed the apparent price of access. NATO confirmed in July 2026 that its communications and information agency had signed the Protected Business Network contract. Accenture estimated its value at approximately €200 million over seven years and said the platform it would deliver with Leonardo was intended to support secure cloud services for approximately 29,000 users across NATO. The United Kingdom separately allocated £100 million through its Defence Investment Plan to the Rapid AI Delivery Taskforce, whose remit covers AI, autonomy and other frontier technologies. These figures buy different things. The NATO contract concerns the classified digital infrastructure on which future services can run. The UK allocation supports adoption across military commands. The budget question is whether each investment produces a defensive task that an operator can complete and verify, with an identifiable cost for keeping that capability available.

A faster finding can produce a longer queue

At Crossed Swords 2026, the NATO Cooperative Cyber Defence Centre of Excellence placed Daybreak and on-premises AI in the exercise environment and introduced a dedicated training stream on AI agents. The exercise put the technology alongside the people who would have to assess its output and coordinate action under time pressure. This is where an apparent gain can reverse. A model can shorten the time needed to identify a possible exposure while increasing the number of cases that require specialist review. If a service owner has limited maintenance windows or lacks authority to make a change, the extra findings accumulate. Procurement should therefore measure the interval from a validated finding to a verified resolution, as well as the interval from first report to validation. The second becomes an advantage only when the first remains manageable and essential services stay available.

National control solves different problems in different countries

France’s August 2026 military programming update places ASGARD, its classified AI supercomputer, within a national investment trajectory that includes renewal of the system. Germany’s Open Defense Cloud addresses a different constraint. BWI says the open-source-based platform is designed to be deployable and to handle classified data, with initial applications for a brigade command post scheduled for 2027. France’s investment gives it an environment for sensitive workloads. Germany’s programme makes the availability of useful data and applications during impaired connectivity a testable requirement. Italy faces a different implementation question. Its 2026–2028 defence planning direction calls for coordination with the National Cybersecurity Agency on crisis management for the defence of the state. In an incident affecting an infrastructure operator and military users, the test is how quickly each authority can verify the facts it needs and identify who has the power to act. Britain’s funded delivery taskforce can contribute a fourth form of evidence by establishing which results from government pilots transfer to military environments with different permissions and service constraints.

Those choices can reinforce one another. The Alliance Digital Strategy, published in January 2026, calls for federated infrastructure and common approaches to data labelling, metadata and access control. Its most valuable implication is that authorised results should be usable across different systems. A French assessment can remain useful to a partner even when the underlying classified material stays in France, provided the recipient can establish its provenance, understand its limits and act within its own authority. A German formation can operate through a loss of connectivity if it retains the information and authority needed for its priority tasks. The protected infrastructure around the model matters as much as the model in both cases. A procurement decision that preserves the ability to retrieve evidence and reconstruct an authorised action gives governments room to replace a supplier or change the way a mission is supported.

Faster response still requires separate legal judgments

Europol’s European Union Terrorism Situation and Trend Report 2026 documents the use of generated and edited media by terrorists and violent extremists. Its attack and arrest series does not isolate AI’s contribution to the disruption of critical systems. That evidentiary limit matters when a serious outage coincides with a claim of responsibility or persuasive synthetic material. Under Directive (EU) 2017/541, specified forms of intentional illegal system or data interference can qualify as terrorist offences when the relevant statutory conditions, including a specified terrorist purpose, are met. An operator can begin restoration as soon as the effects are verified. The competent authorities still need evidence of the conduct and purpose before classifying it. The distinction protects response time and the integrity of any later prosecution.

The same discipline applies to foreign information manipulation. In March 2026, the European External Action Service reported AI-related activity in 147 of the 540 incidents it detected in 2025, while observing limited organic engagement for much of the synthetic material it examined. A crisis responder has a narrower concern than a public-opinion analyst. A disputed instruction can matter if it reaches the people responsible for an emergency decision and consumes the time they have to verify it. The relevant evidence is a chronology linking the message, its authentication and the resulting decision. Articles 4 and 5 of the North Atlantic Treaty leave room for consultation and collective defence under distinct conditions. An authenticated instruction can be issued while responsibility remains under investigation, preserving the ability to respond without getting ahead of the evidence required for escalation.

Europe can agree on the handover before it agrees on the cloud

The 2026 Coordinated Annual Review on Defence, approved on 17 November, identified a European secure defence cloud as an opportunity for cooperation. The most useful early deliverable would be an agreed way to hand over a verified incident. The June 2026 EU–NATO progress report already documents crisis contacts and technical cooperation. The EU–UK Security and Defence Partnership provides a separate route for cooperation on cyber and foreign information manipulation. These channels can be tested against a concrete dependency. ENISA’s Cyber Europe 2026 exercise used interconnected rail and maritime networks to examine consequences for essential services. A more demanding follow-on test would trace whether the civilian operator, the national defence authority and the relevant international counterpart can agree on the service’s condition, accept responsibility for their respective decisions and act within a common operational window. That result would tell ministers which aspects of federation merit funding before a common platform is available.

The next two budget cycles will reveal who carries the risk

Over the next 12–24 months, Germany’s published 2027 Open Defense Cloud milestone and the follow-through to the November European defence review offer two points at which to test the choice governments have made. They have backed national delivery and endorsed wider cooperation. The immediate consequence should be measurable in the time it takes an ally to accept a verified finding, a service owner to close an exposure and a duty team to issue an authenticated instruction when ordinary communications are impaired. Renewal and expansion decisions should rest on those results, including performance when a shared dependency fails.

If governments fund access and defer that work, the cost will fall first on the operators whose services Allied plans assume will be available. Civilian infrastructure providers will carry more of the recovery burden, defence teams will spend scarce time reconciling evidence and authorisations, and finance ministries will face the cost of adapting systems after contracts are in place. The return on the November decisions will become visible when a priority service is restored or a protective action is completed within the time the mission allows, with a record an ally can verify.


Navigational Index

Thematic pillarChapter allocation
AI adoption and military cyber readinessChapter 1. NATO’s adoption baseline and the evidence from Daybreak and Crossed Swords
Chapter 2. Model competition, defence procurement and resilience to provider failure
Cyberterrorism, cognitive warfare and escalationChapter 3. Terrorist use of AI, cyber disruption and the threshold for terrorism
Chapter 4. Foreign information manipulation, decision integrity and crisis escalation
European implementation and choices through 2031Chapter 5. Italy, France, Germany, the United Kingdom and EU–NATO coordination
Chapter 6. Five-year outlook, indicators, courses of action and net assessment

Master Abstract

The military advantage depends on how quickly defenders can close a verified exposure

NATO’s early use of advanced models marks a concrete step in the adoption of AI for cyber defence. OpenAI says it works with the NATO Communications and Information Agency (NCIA) to provide access to NATO’s core institutions, and the exercise organiser independently confirms that Daybreak was used at Crossed Swords alongside on-premises AI. Those records establish two different things. One documents a provider-reported access arrangement and its peak usage. The other documents training in an environment that linked cyber activity with simulated military operations. The next test is whether an analyst can use model output to reach a sound decision faster and whether the responsible team can then make and verify a safe change. If approval, maintenance windows or the availability of engineers determine the pace, faster discovery can enlarge the backlog. If teams can validate and close exposures while maintaining service, the same technology can reduce the time during which military and civilian networks remain vulnerable. That is the operational distinction that adoption figures cannot resolve. Strengthening NATO and allied cyber defenses with AI — OpenAI — Nov 2026. Exercise Crossed Swords tests cyber operations and multi-domain coordination — NATO CCDCOE — Nov 2026. OpenAI

Competition among model providers makes switching a readiness issue

Anthropic’s Project Glasswing, announced in April 2026, and its October expansion of tiered access for qualifying security professionals illustrate how access to advanced cyber capabilities is evolving alongside model performance. OpenAI’s announcement quotes NCIA General Manager Dylan Browne linking usage evidence to cost optimisation and the evaluation of competing solutions. Together, these statements point to a procurement choice between adopting a provider’s integrated workflow quickly and retaining the capacity to switch models without losing logs, permissions or continuity of service. Supporting more than one provider requires additional validation and engineering. It reduces concentration risk only if the alternatives have been tested and can operate through the same disruption. NATO’s revised AI strategy supplies a basis for that approach through its emphasis on traceability, reliability, governability and evaluation. For high-consequence workflows, a repeatable test after a model update or provider outage is more informative than a model ranking at the time of purchase. Project Glasswing: Securing critical software for the AI era — Anthropic — Apr 2026, Oct 2026 update. Strengthening NATO and allied cyber defenses with AI — OpenAI — Nov 2026. Summary of NATO’s revised Artificial Intelligence (AI) strategy — NATO — Jul 2024. anthropic.com

Cyberterrorism and cognitive warfare operate through different mechanisms

Europol documented terrorist and violent-extremist use of generative AI for propaganda and hate speech in its assessment of 2024. ENISA’s assessment of calendar 2025 identifies ransomware, state-linked activity and hacktivist denial-of-service campaigns as distinct parts of the cyber threat landscape. A serious cyber incident can therefore warrant an immediate restoration effort while the actor and motive remain under investigation. Under Article 3 of Directive (EU) 2017/541, specified intentional acts of unlawful system or data interference can qualify as terrorist offences where the applicable statutory conditions are met, including the serious-harm threshold and a specified terrorist purpose. NATO’s Article 5 assessment concerns a separate question. The Alliance assesses case by case whether a significant cyber or hybrid attack amounts to an armed attack. The use of AI or a public claim of responsibility establishes neither the purpose required by the directive nor the effects relevant to collective defence. EU TE-SAT 2025 Executive Summary — Europol — Jun 2025. Exploring the evolution of the cyber threat landscape: How dependencies weaken our digital resilience — ENISA — Sep 2026. Directive (EU) 2017/541 on combating terrorism, Article 3 — European Parliament and Council — Mar 2017. Collective defence and Article 5 — NATO — undated. europol.europa.eu

In this dossier, cognitive warfare refers to attempts to degrade trust, distort situational awareness or influence decisions under pressure. The EEAS evidence concerns the narrower, observable category of foreign information manipulation and interference. Its findings support a judgment that AI is becoming more common in the incidents it detects and that producing multilingual material at scale has become easier. They also show why output and influence must be measured separately. Much of the synthetic content observed in 2025 had little organic engagement. During a real emergency, however, the relevant effect may be a delay in authenticating an instruction or reconciling contradictory reports rather than a durable shift in public opinion. This is a plausible escalation pathway, with the greatest consequence where a decision is time-sensitive and an independent confirmation channel is unavailable. The appropriate measures are time to authenticate, genuine audience exposure and documented decision delay. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — Mar 2026. REPORT ON FIMI THREATS March 2026

European allies enter the transition with different institutions

JurisdictionDocumented baselinePrincipal implementation test
ItalyThe National Cybersecurity Agency’s parliamentary report records 2,729 cyber events handled by CSIRT Italia in 2025, including 615 incidents with confirmed impact. These are response-system figures under Italian definitions. Relazione annuale al Parlamento 2025 — Agenzia per la Cybersicurezza Nazionale — May 2026. 2025Test whether AI helps triage and prioritise verified remediation where civilian service operators and defence dependencies intersect, using confirmed impact and restoration time as outcome measures.
FranceANSSI’s 2025 overview reports that the boundary between state-linked and criminal activity is eroding. OpenAI names ANSSI among the European agencies given access. Access is a provider statement. Panorama de la cybermenace 2025 — ANSSI — Mar 2026. Strengthening NATO and allied cyber defenses with AI — OpenAI — Nov 2026. ANSSITest whether sensitive cases can be referred and corroborated quickly while technical compromise and foreign information interference retain distinct evidentiary standards.
GermanyThe Bundeswehr elevated its Cyber and Information Domain to the fourth armed service in 2024. It also describes a military liaison function inside the national IT situation centre at the Federal Office for Information Security, or BSI. 2022 bis heute: Die Bundeswehr in der Zeitenwende — Bundeswehr — 2026. Status quo: Wo stehen wir in Deutschland in Sachen Cybersicherheit? — Bundeswehr — Feb 2026. bundeswehr.deMeasure whether the military–civilian interface can speed incident coordination when operators hold different access rights and reporting mandates.
United KingdomThe Ministry of Defence’s 2025–26 accounts report a new Defence Cyber and Electromagnetic Force. The NCSC separately assesses a widening divide between systems that can keep pace with AI-enabled threats and more vulnerable systems through 2027. Ministry of Defence Annual Report and Accounts 2025–26 — UK Ministry of Defence — Jul 2026. Impact of AI on cyber threat from now to 2027 — NCSC — May 2025. 26Test continuity across military and civilian dependencies when AI services or communications are degraded. Treat the NCSC’s 2027 forecast as a near-term assessment with its own horizon.
European UnionThe Cyber Solidarity Act entered into force on 4 February 2025 and provides for cross-border cyber hubs and an emergency mechanism that includes the EU Cybersecurity Reserve. Article 2(3) of the AI Act excludes systems where and insofar as they are used exclusively for military, defence or national security purposes. EU Cyber Solidarity Act — European Commission — undated. Article 2: Scope — European Commission AI Act Service Desk — Jul 2026 consolidated text. Shaping Europe’s digital futureAgree what findings, safeguards and escalation records can travel between civilian and military authorities. Assess AI Act scope against the actual purpose of each deployment, especially where infrastructure serves both uses.

Key Evidence Table

IndicatorValue/statusReference dateDefinition/scopeIssuerExact source
NATO model usageUp to 300 billion tokens per month, provider-reportedDisclosed 17 Nov 2026. Peak month undisclosedRaw input and output tokens combined in the reported OpenAI–NCIA work. Workload breakdown, spend and outcome data are undisclosed in the announcement.OpenAIStrengthening NATO and allied cyber defenses with AI — OpenAI — Nov 2026. OpenAI
Exercise integrationMore than 250 participants from over 40 countries. New AI-agent training stream2–6 Nov 2026Crossed Swords execution phase in Tallinn. Exercise-wide participation across NATO and non-NATO countries, with on-premises AI and Daybreak used in the technical environment.NATO CCDCOEExercise Crossed Swords tests cyber operations and multi-domain coordination — NATO CCDCOE — Nov 2026. ccdcoe.org
AI-related foreign information manipulation147 of 540 detected incidents, reported as 27%, versus 41 AI-related cases in 2024Calendar 2025, with 2024 comparisonIncidents involving AI-related tactics, techniques or procedures within EEAS monitoring coverage. Changes in detection and coverage can affect comparisons.EEAS4th EEAS Report on Foreign Information Manipulation and Interference Threats, 2025 FIMI in Numbers and AI-Enhanced FIMI Activities — EEAS — Mar 2026. REPORT ON FIMI THREATS March 2026

Competing Pathways Through 2031

Defensive gains become repeatable where institutions combine model access with skilled review, authority to implement fixes and continuity under degraded conditions. This is the pathway with the clearest documented starting point. Its strongest confirmation would be lower time from verified finding to deployed fix and faster recovery, sustained across model versions and comparable incidents.

Threat volume outruns remediation where attackers benefit from more capable tools while defenders cannot validate findings, maintain legacy systems or implement changes quickly enough. The NCSC’s near-term assessment supports this mechanism. The clearest signal would be faster discovery paired with a growing backlog of confirmed high-priority exposures and repeated disruption at poorly resourced operators. Impact of AI on cyber threat from now to 2027 — NCSC — May 2025. National Cyber Security Centre

Information interference imposes its greatest cost during crises where a genuine service interruption or security incident already strains authoritative communications. The decisive indicator would be independently documented confusion or delay in consequential decisions. More synthetic material, by itself, provides weak evidence for this pathway. The three pathways can coexist across sectors or during the same crisis. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — Mar 2026. REPORT ON FIMI THREATS March 2026

Principal Gaps and Watch Indicators

Consequential gapRecord or observation that would resolve itDecision threshold
NATO’s security return on model usageA public NCIA evaluation or procurement record linking comparable workloads to verified fixes, recovery time, analyst workload and total cost. The vendor announcement provides the usage figure. OpenAIExpand high-consequence deployment when repeat testing shows improved outcomes against the existing workflow and exercises demonstrate recovery after model or provider failure.
Exercise performance under degraded conditionsA CCDCOE after-action account with comparable AI-assisted and baseline results, including false escalations, operator overrides and continuity when AI is unavailable. The public exercise account describes the training design. ccdcoe.orgRetain human authorisation for consequential operational changes wherever review, auditability or rollback fails under exercise conditions.
Transfer from model-assisted findings to durable fixesThe primary ENISA–CERT‑EU evaluation, including case selection, denominators and the disposition of findings. OpenAI summarises the study, but its primary text could not be independently inspected for this assessment. OpenAIUse study results in deployment decisions once the evaluation population, verification process and transferability to the intended environment can be checked.
AI’s contribution to terrorist cyber disruption and cognitive effectsJudicial or competent-authority findings establishing intent and harm for terrorist cyber incidents, alongside incident reviews showing that manipulated information materially delayed authentication, response or public instructions. eur-lex.europa.euReassess the threat balance when competent records establish consequential effects attributable to AI use, with a clear account of the role it played.

AI in NATO Defence

Evidence as of 18 November 2026. The interactive charts and five-language selector require JavaScript.

Selected evidence and its scope
IndicatorValueScope and source
NATO model usage Up to 300 billion tokens per month Raw input and output tokens combined. Provider-reported peak usage. OpenAI, 17 November 2026
Crossed Swords 2026 More than 250 participants from over 40 countries Exercise-wide participation, Tallinn, 2–6 November 2026. NATO CCDCOE
AI-related foreign information manipulation 147 of 540 detected incidents, reported as 27% EEAS monitoring sample for 2025. EEAS, March 2026
CSIRT Italia annual counts. ACN parliamentary report, Figure 2
YearCyber eventsConfirmed incidentsIncident notifications
20231,411303349
20241,979573450
20252,729615661

Pillar I — AI Adoption and Military Cyber Readiness

Chapter 1. NATO’s Adoption Baseline and the Evidence from Daybreak and Crossed Swords

NATO now has a public record of advanced AI access and use in cyber exercises. The next readiness decision is whether AI-assisted teams can complete more verified defensive work within the same operational window, while preserving service continuity and clear authority to act. The distinction matters because the Alliance has three forms of evidence in view. OpenAI’s November announcement describes access through the NATO Communications and Information Agency (NCIA) and reports peak consumption. The NATO Cooperative Cyber Defence Centre of Excellence (CCDCOE) independently documents use of Daybreak and on-premises AI in an exercise environment. A national responder, CERT Polska, separately documents model involvement in research that led to an actual vulnerability advisory and vendor fixes. These records illuminate different stages of adoption. A usage figure establishes the scale claimed by the provider, an exercise establishes integration into training, and a published advisory establishes a real defensive outcome in a specific case. None supplies a matched comparison of NATO teams operating with and without AI. The case for widening deployment therefore rests on collecting that comparison at the level of the defensive workflow. Strengthening NATO and allied cyber defenses with AI — OpenAI — November 2026. Exercise Crossed Swords tests cyber operations and multi-domain coordination — NATO CCDCOE — November 2026. Critical vulnerabilities in MikroTik RouterOS are being actively exploited. Immediate update recommended — CERT Polska — September 2026. OpenAI

NATO already has a timetable for moving from demand to integration

The Rapid Adoption Action Plan, endorsed by Allied leaders on 25 June 2025, gives NATO and its members a practical framework for evaluating commercial AI without allowing a single supplier’s product cycle to dictate the timetable. The plan generally aims for 24 months from identification of a need through acquisition and integration into Allied armed forces. It envisages around three months for market research and, once potential solutions have been identified, generally 12 months for incremental testing, evaluation, verification, validation and integration. The figures are policy objectives with distinct starting points. The plan gives no basis for adding them into a contractual deadline or assuming that a subsequent model release inherits an earlier version’s acceptance. For cyber defence, the implication is that procurement can move at different speeds for different levels of authority. A tool that organises an approved body of evidence can enter a bounded trial relatively quickly. A tool that informs a consequential change on a live system requires evidence about what happens when the input is incomplete, the recommendation is wrong or the service must continue through a failed change. The same procurement can therefore advance quickly while reserving wider operational use for a later decision. Summary of NATO’s Rapid Adoption Action Plan, paragraphs 8–9 — NATO — June 2025. NATO Official text

The Alliance has facilities for collecting part of that evidence before a product reaches a live network. In its July 2026 account, NATO described the Defence Innovation Accelerator for the North Atlantic (DIANA) as having 16 affiliated accelerator sites and more than 200 test centres. NATO identifies those test centres as laboratories, testbeds and field environments available for testing and validation by selected innovators. Their value for Allied cyber adoption would come from testing a narrowly defined defensive task against a common evidence standard, then establishing which parts of the result transfer to a second organisation. A tool that performs well with complete inventories, modern telemetry and experienced reviewers may offer less benefit where asset records are incomplete or teams lack authority to remediate. A useful trial would distinguish limitations of the model from limitations of the environment by recording what information was available, which operator checked the recommendation and where the workflow stopped. The test network provides venues for such work. Acceptance for a particular mission environment still depends on that environment’s data, permissions, service dependencies and accountable authority. Defence Innovation Accelerator for the North Atlantic (DIANA) — NATO — July 2026 update. NATO Topic

Crossed Swords and Locked Shields expose different readiness questions

Crossed Swords is most informative about how AI fits into an operational chain. The organiser’s account places the new AI-agent training stream alongside established work on coordination between military command elements, technical specialists and civilian partners concerned with critical infrastructure. That structure makes authority and timing central to the assessment. An exercise can establish that operators received and examined an AI-assisted recommendation, then ask whether it arrived while they still had time to verify it and whether the designated authority could act on it. A separate exercise offers a different vantage point. CCDCOE reported more than 4,000 participants from 41 nations at Locked Shields 2026, which concluded on 24 April and focused on protecting critical infrastructure and military systems under simulated pressure. Its size illustrates the breadth of the environment in which civilian and military teams can practise continuity and coordination. Participation in either exercise is a measure of scale, and the exercises have different designs. Their complementary value lies in testing distinct questions. Crossed Swords can reveal where an AI-supported recommendation fits into a time-sensitive coordination process. A continuity exercise can reveal whether essential services remain supportable when analysts lose that assistance or have to reconcile conflicting information. Exercise Crossed Swords tests cyber operations and multi-domain coordination — NATO CCDCOE — November 2026. Locked Shields 2026 united the power of 41 nations to defend cyberspace — NATO CCDCOE — April 2026. ccdcoe.org

The CERT Polska advisory improves the public evidence in a different way. In September 2026, the national responder named the researcher and the OpenAI models used in the work, published its assessment of the affected product and directed operators to available vendor fixes. The institutional significance is the existence of a documented path from AI-assisted research to an accountable public advisory and a remediation decision by the operator. The public record cannot establish how much time the models saved or how quickly affected organisations installed the fixes. That gap identifies two separate readiness measures for NATO. One is the time a team needs to produce a verified finding that another analyst can reproduce. The other is the time the responsible service owner needs to apply and verify a fix. If the first interval shortens while the second expands, the organisation has improved its ability to detect problems while leaving exposures open longer. A serious deployment review should record both intervals, along with the age of the remaining backlog and any adverse effects on service. Critical vulnerabilities in MikroTik RouterOS are being actively exploited. Immediate update recommended — CERT Polska — September 2026. CERT Polska

What the adoption record establishes

Public recordVerified figure or observationWhat it establishesDecision it can inform
Rapid Adoption Action Plan24 months general adoption objective, 12 months generally for incremental testing and integration after identifying potential solutions, and three months generally for market researchAn Allied timetable for moving from operational need towards acquisition and integrationSet separate milestones for market review, limited trial and operational acceptance. NATO official text — June 2025. NATO Official text
DIANA network16 affiliated accelerator sites and more than 200 test centres as described in July 2026Access to a distributed network of facilities for selected innovatorsSeek replication across environments before relying on a single test result. NATO DIANA overview — July 2026. NATO Topic
Crossed Swords 2026Daybreak and on-premises AI were used in the technical environment, with a dedicated agent training streamOrganiser-confirmed exercise integration and an emphasis on operator oversightRequest evidence from comparable AI-assisted and existing workflows before expanding authority. CCDCOE exercise account — November 2026. ccdcoe.org
Locked Shields 2026More than 4,000 participants from 41 nationsScale of a separate defensive training environment covering critical infrastructure and military systemsUse continuity and coordination outcomes for evaluation. Treat attendance as participation data. CCDCOE concluding account — April 2026. ccdcoe.org
CERT Polska advisoryThe responder publicly attributed use of two specified models in its research and identified available vendor fixesA documented national defensive case linking researcher oversight, model assistance and public remediation guidanceAudit how findings were verified and how long remediation took in comparable cases. CERT Polska advisory — September 2026. CERT Polska

Key judgments. The strongest near-term case is for AI assistance at points where the output can be checked against independent evidence, the responsible operator can reject it and the service owner can complete and verify the action. A configuration that generates more findings without adding capacity to review or resolve them can make prioritisation harder. The right unit of comparison is a completed defensive task under defined conditions. This gives decision-makers a common basis for judging different tools even where national systems, clearance rules or service dependencies differ.

What would change the assessment. An exercise report showing repeatable improvements in verified remediation and service restoration, with unchanged or lower rates of harmful recommendations and unnecessary escalations, would justify greater confidence. If gains consistently disappear when information is incomplete, an approval is delayed or the primary service is unavailable, the evidence favours retaining AI in advisory roles for those workflows.

Open official record. The Crossed Swords account describes the exercise design and technology. It provides no matched outcome data by workflow or deployment environment. Public evidence of readiness would be strengthened by a CCDCOE after-action report identifying the task, baseline, reviewer workload, continuity requirement and disposition of consequential recommendations. Exercise Crossed Swords tests cyber operations and multi-domain coordination — NATO CCDCOE — November 2026. ccdcoe.org

Chapter 2. Model Competition, Defence Procurement and Resilience to Provider Failure

The most consequential procurement decision is which parts of the defensive workflow NATO and its members can carry across providers and which parts must remain usable when every approved model is unavailable. The Alliance Digital Strategy, published in January 2026, sets the institutional context for that decision. It calls for federated platforms, a hybrid cloud model spanning multiple classification levels, and common approaches to data labelling, metadata and access control. Applied to defensive AI, these principles imply that a supplier’s interface should sit inside an institutional workflow whose underlying evidence, permissions and decision records remain intelligible to the people responsible for the mission. If a service changes its model, an analyst must still be able to identify the evidence available at the time, who signed off and which systems were affected. If the service becomes unavailable, the operations team still needs the approved record and a way to act on it. Passing the same prompt to a second provider proves little if the alternative lacks access to the authorised evidence or loses the ability to reconstruct the prior decision. Alliance Digital Strategy — NATO — January 2026. Summary of NATO’s revised Artificial Intelligence (AI) strategy — NATO — July 2024. NATO Official text

The Protected Business Network shows the scale of the enabling infrastructure

On 7 July 2026, NATO confirmed that NCIA had signed a contract with Accenture for the Protected Business Network (PBN), describing it as a foundation for classified digital operations across NATO. Accenture estimated the contract at approximately €200 million over seven years and stated that it and Leonardo would design, implement and operate the core platform across a multi-cloud environment provided by NCIA, supporting the progressive adoption of secure cloud services by approximately 29,000 users. The contract value is a supplier estimate and the user figure is prospective. They describe the infrastructure programme and tell decision-makers nothing directly about expenditure on Daybreak, the number of AI users or which models are approved for classified workloads. The relevance of the award lies in how much military AI depends on the infrastructure around it. A model can be replaced more readily when evidence remains retrievable, access controls travel with the workflow and operator tools preserve a common record. Conversely, a programme can use several clouds while retaining a common point of failure in identity, networking or evidence storage. The readiness test is whether a specifically named dependency can fail while the relevant defensive task continues. NATO builds a more agile and resilient digital infrastructure — NATO — July 2026. NATO Announces Major Contract with Accenture to Help Advance Towards a More Agile and Resilient Digital Infrastructure — Accenture — July 2026. NATO News

Competition extends to access terms, updates and exit costs

In October 2026 Anthropic announced three access tiers in its expanded Cyber Verification Program and said existing Glasswing members would move to the specialised tier. The commercial significance is that a buyer’s capacity can depend on continued eligibility and the conditions attached to a particular service, in addition to its technical performance. A tender should therefore ask what happens if a model is withdrawn, a capability is restricted, a version changes or the buyer needs to move approved work to another environment. The UK government’s cloud guidance recognises the potential operating benefits of managed services and recommends monitoring where switching costs grow disproportionately. The UK National Cyber Security Centre makes the complementary point that changes to data, models or prompts can alter system behaviour and that operational teams need testing and monitoring around updates. Together, these sources support treating material model changes as acceptance events for the affected workflow. Where a supplier offers versioned access, the operator can compare the replacement against an approved baseline. Where the supplier controls the release schedule, a tested suspension and fallback procedure becomes more valuable because the operator has less control over exactly when behaviour changes. Project Glasswing: Securing critical software for the AI era — Anthropic — October 2026 update. Managing technical lock-in in the cloud — UK Government Digital Service and Central Digital and Data Office — September 2026 update. Guidelines for secure AI system development, Secure operation and maintenance — UK NCSC — 2023. anthropic.com

A procurement evaluation should put the strongest model and the best operating arrangement through the same mission test. First, establish the value delivered by the current workflow with the authorised data, operator time and approval chain actually available. Then evaluate competing arrangements against the same task and evidence, including a period in which the primary service is deliberately unavailable. Record both the cost of running the workflow and the cost of keeping it ready. A model with inexpensive inference may require extensive human reconciliation. A second commercially hosted model may offer excellent everyday switchability while relying on the same identity or connectivity layer. A locally operated service may offer more control over when it is updated while increasing the staffing, assurance and infrastructure burden. For the most critical tasks, a manual workflow can preserve authority and access to evidence through broader failure, with less throughput. The commercial comparison should therefore price human review, integration, assurance and rehearsed transition alongside service consumption. No public NATO workload or contract breakdown reviewed for this chapter supports a monetary estimate for those costs. Managing technical lock-in in the cloud — UK Government Digital Service and Central Digital and Data Office — September 2026 update. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — NIST — July 2024. GOV.UK

Procurement options and their consequences

These are proposed operating arrangements and relative assessments. Actual authority, acceptance criteria and recovery targets belong to the contracting organisation and the relevant mission owner.

Course of action and authorityExpected effectBurden and time to effectReversibilityPrincipal risk and second-order consequence
One primary supplier with a tested manual workflow. The contracting authority purchases access. The mission owner and security authority approve its scope.Integrates AI assistance into a bounded workflow and preserves manual continuity.Lower initial integration burden. Time to effect depends on access authorisation and operator training.High where evidence, decisions and procedures remain portable. Decreases if proprietary tooling becomes central to the workflow.Human capacity can erode when everyday work transfers to the service. A prolonged outage can turn a routine fallback into a staffing problem.
Primary supplier plus an independently tested alternative. The contracting and security authorities approve both services. The mission owner sets the continuity requirement.Supports a switch after provider failure, version withdrawal or a change in access conditions.Higher integration, validation and training burden. Benefits begin once the alternative has passed a representative exercise.Higher when both services can use the necessary approved evidence and keep an interoperable record.A shared identity, network or storage dependency can disable both services. Duplicate oversight also competes for scarce staff.
Locally controlled inference for a bounded mission. The operating organisation accepts responsibility for infrastructure and ongoing assurance. The security authority approves the deployment.Reduces reliance on external inference availability for the approved task and gives the operator greater control over updates.Greater upfront and continuing burden for compute, security testing and specialist staff. Time to effect depends on accreditation and support capacity.Moderate. The operator controls transition timing but carries infrastructure and staffing commitments.A local deployment may become harder to maintain or fall behind newer services. Its value depends on having a tested process for evaluating and installing updates.

Defensive AI readiness and service continuity

Proposed acceptance sequence and fallback decisions for authorised defensive workflows.

From a finding to a verified outcome

  1. 1. Establish the caseRecord the defensive task, approved evidence and responsible service owner.
  2. 2. Verify the findingCheck the result against independent evidence. Hold unresolved findings for review.
  3. 3. Authorise the changeThe designated owner records approval, continuity requirements and a reversal plan.
  4. 4. Confirm the outcomeCheck that the exposure is closed and the service is stable. Measure review and remediation time.

When the primary service is impaired

Check whether the alternative can access authorised evidence, enforce the required permissions and preserve the decision record.

  • Alternative acceptedSwitch to the approved service and confirm continuity before restoring its full role.
  • Alternative outside its approved scopeContinue in the manual workflow for priority cases and defer work that cannot be verified.
  • Evidence or authority unavailableUse the service continuity procedure and suspend changes that require an unavailable approval or record.
Operational acceptance table
Proposed evidence for each readiness decision
DecisionEvidence to retainCondition for proceeding
Expand a trialMatched baseline and assisted case records, reviewer time and confirmed outcomes.Repeated improvement in completed defensive work within the service’s tolerance for error.
Accept a model updateVersion record, repeat evaluation and results for previously approved cases.The updated service meets the workflow’s existing acceptance criteria.
Rely on an alternativeTimed switch, dependency check, permission review and reconstructed decision history.The alternative completes the priority task within the mission owner’s recovery limit.
Resume after disruptionApproved service status, reconciliation of changes made during fallback and a record of unresolved cases.The service owner confirms continuity and the operating authority approves resumption.

Reference documents — NATO Alliance Digital Strategy, NATO Rapid Adoption Action Plan, UK NCSC secure operation and maintenance guidance.

The measures that should govern renewal

A contract renewal should turn on whether the arrangement improves completed defensive work at a sustainable cost and continues to support the mission during disruption. For each approved workflow, the acceptance record should preserve performance under normal conditions, after a material model update, during primary-provider loss and during any shared-dependency failure that the mission owner considers consequential. The comparison should include human review time, confirmed closure, unintended service disruption, backlog age and the time required to restore the priority defensive task. The authority responsible for the service should define its maximum tolerated interruption before the test. An alternative that restores access after that point may still be valuable for routine work, but it cannot satisfy that mission’s continuity requirement. The same exercise should establish what happens to decisions made during fallback and how they are reconciled when the primary service returns. This is where continuity becomes an operational capability rather than a statement about vendor diversity. Alliance Digital Strategy — NATO — January 2026. Guidelines for secure AI system development, Secure operation and maintenance — UK NCSC — 2023. NATO Official text

Key judgments. A single provider can be a defensible choice for a bounded workflow with a reversible implementation and a capable manual fallback. A second provider earns its cost when it restores a priority function through a failure the primary provider cannot survive. Local operation earns its cost where external availability, handling requirements or control of updates make a hosted service insufficient for the mission and the organisation can sustain the local capability.

What would change the assessment. Demonstrated continuity across an exercised provider failure, with the relevant permissions, evidence and decision history intact, would support greater reliance on AI in higher-consequence workflows. If a supposedly independent alternative fails under the same dependency loss, the concentration risk remains and the case for funding an additional supplier weakens until the shared dependency is addressed.

Open official record. NATO’s public PBN announcement establishes the infrastructure award. The open Daybreak record establishes provider-reported access and usage. The evidence reviewed leaves the pricing structure, workload-level service terms, material model-change commitments, export provisions and tested recovery arrangements for the reported NCIA work unresolved. A public contract notice or assurance summary addressing those points would materially improve the procurement assessment. NATO builds a more agile and resilient digital infrastructure — NATO — July 2026. Strengthening NATO and allied cyber defenses with AI — OpenAI — November 2026. NATO News


Pillar II — Cyberterrorism, Cognitive Warfare and Escalation

Chapter 3. Terrorist Use of AI, Cyber Disruption and the Threshold for Terrorism

The clearest documented effect of generative AI on terrorism is its use in producing and adapting propaganda. The more consequential proposition, that AI is materially increasing terrorist disruption of critical systems, remains unproven in the public evidence reviewed. Europol’s 2026 assessment documents terrorist and violent-extremist use of image, audio and video generation in the online environment. Its aggregate statistics on attacks and arrests do not isolate the operational contribution of AI. The distinction matters for resource allocation. Investigators need to determine whether a tool was present, whether it materially assisted the conduct under investigation and whether the conduct meets the elements of a terrorism offence. An infrastructure operator has a different immediate responsibility. It must establish what services are impaired, safeguard continuity and preserve evidence while motive remains unresolved. A government can therefore act rapidly on a confirmed disruption while reserving judgment on the actor, the role of AI and the legal classification. This sequencing becomes particularly important when an online claim of responsibility circulates before the operational facts are established. European Union Terrorism Situation and Trend Report 2026, pp. 10–12 and 17–19 — Europol — July 2026. europol.europa.eu

The terrorism figures establish the baseline for assessing AI’s contribution

Europol recorded 45 reported terrorist attacks across 10 EU Member States in 2025, comprising 22 completed, 20 foiled and three failed attacks. Completed attacks caused six fatalities. In 2024, 14 Member States reported 58 attacks, of which 34 were completed, causing five fatalities. Arrests moved in the other direction. Europol reports 486 suspects arrested for terrorism-related offences in 21 Member States in 2025, against 449 in 2024 and 426 in 2023. Calculated from the published counts, reported attacks fell 22.4% from 2024 to 2025, while recorded arrests rose 8.2%, both rounded to one decimal place. These changes describe different populations and processes. An arrest can concern an offence or investigation outside the reported attack series. The count of foiled attacks also depends on which plots authorities detect, investigate and report. The figures therefore cannot establish a common rate of attack prevention or show whether AI made more plots possible. A credible estimate of AI’s contribution would require case-level evidence connecting its use to a defined step in the offence, with a clear account of the other resources available to the actor. European Union Terrorism Situation and Trend Report 2026, p. 18 — Europol — July 2026. europol.europa.eu

The strongest evidence of AI use lies further upstream. Europol describes synthetic and edited material circulating in extremist environments and observes that a substantial part of what it saw consisted of hate speech or material that was offensive but lawful, alongside content that warranted a terrorism assessment. The United Nations Office of Counter-Terrorism documents examples including a synthetic news presenter used to disseminate Da’esh propaganda and AI-generated imagery used to reinforce a far-right conspiracy narrative. Both cases show how generative tools can lower the cost of producing material in formats an audience already uses. Their downstream effect depends on whether real audiences see the material, whether it changes behaviour and whether any resulting conduct crosses a legal threshold. This distinction also applies to platform response. Europol reports that competent national authorities and Europol assessed 81,862 pieces of online content under its terrorism heading in 2025, leading to 61,412 referrals to 99 hosting service providers. These are assessments and referrals within a defined institutional workflow. They provide no count of synthetic items, unique viewers, completed removals or people recruited. Treating any one of those outcomes as equivalent to the volume of referrals would obscure where a counter-terrorism intervention had an actual effect. European Union Terrorism Situation and Trend Report 2026, pp. 11–12 and p. 11 note 13 — Europol — July 2026. Practice Guide on Artificial Intelligence and Preventing and Countering Violent Extremism — United Nations Office of Counter-Terrorism — 2026. europol.europa.eu

IndicatorVerified figureUnit and reference periodWhat the figure can support
Reported terrorist attacks45 in 2025, compared with 58 in 2024Completed, failed and foiled attacks. Reported by 10 Member States in 2025 and 14 in 2024.A comparison of reported attack totals, with the underlying reporting coverage stated. EU TE-SAT 2026, p. 18 — Europol. europol.europa.eu
Status of attacks reported for 202522 completed, 20 foiled, three failedStatus assigned to the 45 attacks. Six fatalities resulted from completed attacks.A distinction between completed harm and reported plots that failed or were disrupted. EU TE-SAT 2026, p. 18 — Europol. europol.europa.eu
Terrorism-related arrests426 in 2023, 449 in 2024, 486 in 2025Suspects arrested. The 2025 total covers 21 Member States.A trend in recorded enforcement activity under the report’s definitions. EU TE-SAT 2026, p. 18 — Europol. europol.europa.eu
Online assessment and referral activity81,862 items assessed, 61,412 referrals to 99 hosting service providersCalendar 2025, under the terrorism heading, for competent Member State authorities and Europol.The scale of assessment and referral activity. Removal, audience and AI-use data require separate records. EU TE-SAT 2026, p. 11 note 13 — Europol. europol.europa.eu

The legal threshold depends on the act, its gravity and its purpose

Article 3 of Directive (EU) 2017/541 provides the relevant EU criminal-law framework, implemented through national law. Its opening test covers intentional acts defined as offences under national law that, by their nature or context, may seriously damage a country or an international organisation, when committed for one of three specified purposes. Those purposes are serious intimidation of a population, undue compulsion of a government or international organisation, or serious destabilisation or destruction of fundamental political, constitutional, economic or social structures. For illegal system interference, Article 3(1)(i) refers to the circumstances in Article 9(3) or Article 9(4)(b) or (c) of Directive 2013/40/EU. These cover a significant number of affected systems in the specified tool-related circumstances, serious damage, or conduct against a critical-infrastructure information system. For illegal data interference under the same terrorism provision, the cross-reference is narrower and covers the critical-infrastructure circumstance in Article 9(4)(c). The cross-referenced circumstance is only part of the test. A serious outage at a critical operator supplies an urgent basis for response and may satisfy one relevant condition. A terrorism determination still requires evidence of the applicable offence and statutory purpose. The use of an AI model does not change those elements. Directive (EU) 2017/541 on combating terrorism, Article 3 — European Parliament and Council — March 2017. Directive 2013/40/EU on attacks against information systems, Articles 4, 5 and 9 — European Parliament and Council — August 2013. eur-lex.europa.eu

Online content presents a related but separate legal question. Article 5 of Directive (EU) 2017/541 addresses public provocation to commit a terrorist offence. It requires, among other elements, intent to incite and conduct that creates a danger that one or more of the relevant offences may be committed. Regulation (EU) 2021/784 separately establishes the removal-order framework for terrorist content online, including the one-hour time limit after a hosting service provider receives a competent authority’s removal order. That procedural deadline concerns action on the order. Europol’s referrals statistic measures a different administrative activity. For the analyst, the useful distinction is between provenance, purpose and unlawful conduct. A synthetic-media assessment can help determine how a recording was produced. The text, surrounding conduct and legal context bear on whether publication meets the applicable threshold. Separating these judgments gives authorities room to act promptly on a lawful order while retaining the evidence required for review and ensuring that the presence of AI does not displace the substantive test. Directive (EU) 2017/541 on combating terrorism, Article 5 — European Parliament and Council — March 2017. Regulation (EU) 2021/784 on addressing the dissemination of terrorist content online — European Commission overview — 2021 instrument. eur-lex.europa.eu

Key judgments. AI already affects the production and adaptation of extremist media. Its contribution to serious terrorist cyber disruption remains a case-level question. The circumstances of a disruption, evidence of unlawful conduct and evidence of a specified terrorist purpose should each receive a separate finding. Technical responders can protect services and preserve evidence while investigators resolve the latter two questions.

What would change the assessment. A documented investigation or judgment establishing both a qualifying cyber offence and a material AI contribution would strengthen the case for treating AI-assisted terrorist disruption as a demonstrated operational threat. Several comparable findings would allow policymakers to assess whether the contribution recurs across actors, sectors or types of incident.

Open official record. Europol’s public series provides terrorism counts and qualitative findings on AI use. The decisive collection gap is a case-level record that identifies the contribution of AI, the outcome it helped produce and the competent authority’s assessment of the offence and purpose. European Union Terrorism Situation and Trend Report 2026 — Europol — July 2026. europol.europa.eu

Chapter 4. Foreign Information Manipulation, Decision Integrity and Crisis Escalation

AI’s most consequential effect in cognitive warfare may be to consume the time available to verify evidence and issue instructions during a real crisis. An information operation can matter even when it attracts little sustained public engagement, provided it interferes with a narrow audience that must make a consequential decision within a short window. That pathway remains an analytical judgment. The European External Action Service (EEAS) documents the infrastructure of foreign information manipulation and interference, including coordinated activity across channels and the use of synthetic content. It also reports that much of the AI-generated material it observed in 2025 attracted little organic engagement. Those findings place a limit on claims of broad persuasion while leaving a narrower operational question open. Did an incident commander, public authority or infrastructure operator take longer to establish which instruction was genuine, and did that interval affect the response? The answer requires time-stamped operational evidence. An impression count cannot show when an authorised decision-maker learned of a claim, what evidence was available at that moment or whether an independently authenticated instruction would have changed the sequence of events. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — March 2026. REPORT ON FIMI THREATS March 2026

The EEAS evidence reveals infrastructure and the limits of attribution

The EEAS identified 10,500 unique channels involved in the foreign information manipulation incidents it detected in 2025. These included websites, social media accounts and blogs. Its network visualisations drew on approximately 3,000 of the most recurrent assets. Within that selected subset, the EEAS reported 9.5% as directly related to a state actor through official or state-controlled channels and 90.5% as covert assets linked to or aligned with a threat actor under its classification. The sample was selected for recurrence, so those proportions should remain attached to the mapped subset. The report also states that 65% of recorded incidents remained unattributed, even though it identified indicators of coordination with assets linked to Russian or Chinese attributed infrastructures. This is a substantive distinction for a crisis cell. The identity of a channel, its relationship to another account and the responsibility of a state or organisation for the overall incident are separate findings. Evidence for one can increase confidence in another, but it cannot by itself establish who authorised the operation. Institutional assessments should preserve the chain of inference so that remedial action, public attribution and diplomatic escalation can each rely on the evidence appropriate to that decision. 4th EEAS Report on Foreign Information Manipulation and Interference Threats, 2025 FIMI in Numbers and network analysis — EEAS — March 2026. REPORT ON FIMI THREATS March 2026

France’s Rokh Solis investigation shows why asset mapping and audience effect should be recorded separately. In its technical report of 11 June 2026, VIGINUM described an operation directed at La France Insoumise and candidates in the March municipal elections, including in Marseille, Toulouse and Roubaix. It identified four websites with common technical characteristics and coordinated relay activity on TikTok, Instagram, X and Facebook, and linked two additional websites to the same ecosystem. VIGINUM concluded that the operation met its criteria for foreign digital interference. It also found that the operation achieved low online visibility before the election despite repeated attempts at artificial amplification. This makes the case useful as a check on inflated impact estimates. There was evidence of coordination and deceptive activity across several channels, and the public record still supplies no measured change in voter behaviour attributable to the campaign. The case also exposes a practical choice for authorities. They can preserve evidence, notify the relevant institutions and investigate foreign involvement without publicly elevating material whose circulation remains limited. The appropriate point for a wider public correction depends on whether affected citizens need an authoritative instruction or clarification to act. Rokh Solis : Analyse d’un mode opératoire informationnel ayant ciblé les élections municipales de mars 2026 — VIGINUM/SGDSN — June 2026. sgdsn.gouv.fr

Public evidenceMeasurement boundaryDecision implication
10,500 unique channels associated with EEAS-detected incidents in 2025A count of observed channels, including multiple types of asset. It does not measure the size of an audience.Use the network evidence to prioritise corroboration and avoid counting multiple relays as multiple independently verified events. Fourth FIMI threat report — EEAS — March 2026. REPORT ON FIMI THREATS March 2026
Approximately 3,000 recurrent assets mapped. 9.5% directly state-related and 90.5% covert in the selected subset.Channel-level classification in a recurrence-selected subset. The categories use the EEAS definitions.Preserve the distinction between a mapped relationship and evidence of responsibility for a specific operation. Fourth FIMI threat report — EEAS — March 2026. REPORT ON FIMI THREATS March 2026
65% of incidents remained unattributed in the EEAS sampleIncident-level finding. Indicators of coordination were recorded, with actor responsibility still unresolved in these cases.Communicate confidence separately for coordination and actor attribution. Fourth FIMI threat report — EEAS — March 2026. REPORT ON FIMI THREATS March 2026
Four technically related sites, two additional linked sites and low pre-election visibility in Rokh SolisFindings on one operation documented by VIGINUM. The technical report describes activity around the March 2026 French municipal elections.Evaluate reach and documented effects independently of the complexity of the operation. Rokh Solis technical report — VIGINUM/SGDSN — June 2026. sgdsn.gouv.fr

Crisis decisions require separate thresholds for authentication, attribution and escalation

The most useful crisis record would capture when the disputed material first reached the responsible decision-maker, when the underlying event was verified, when an authoritative instruction was issued and what delay, if any, the false material imposed. Authentication can proceed while attribution remains open. A service owner can confirm the actual operating condition and an authorised spokesperson can communicate verified instructions even while investigators assess the origin of a disputed recording. The EU AI Act addresses a different part of the problem through Article 50(4), which sets disclosure requirements for deployers who generate or manipulate image, audio or video constituting a deep fake, subject to the provision’s qualifications. A disclosure duty can improve transparency where it applies. In an actual crisis, the decision-maker still needs an independently authenticated instruction and a record of who issued it. Absence of a label establishes neither authenticity nor attribution. The operational metric is whether the responsible person obtained trustworthy evidence in time to make the required decision. Regulation (EU) 2024/1689, Article 50(4), consolidated text — European Parliament and Council — July 2026 version. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — March 2026. eur-lex.europa.eu

European and Allied institutions also have distinct escalation instruments. On 15 December 2025, the Council of the EU announced restrictive measures against 12 individuals and two entities in connection with Russia’s hybrid activities, including information manipulation and malicious cyber activity. Council Decision (CFSP) 2026/2258, adopted on 8 October 2026, extended the relevant framework until 9 October 2027. These are documented uses of EU restrictive measures under a specific legal regime. Under the North Atlantic Treaty, Article 4 provides for consultation when an Ally considers that the territorial integrity, political independence or security of a party is threatened. NATO’s Vilnius communiqué separately states that a single or cumulative set of malicious cyber activities could reach the level of armed attack and lead the North Atlantic Council to invoke Article 5 on a case-by-case basis. Where incidents are assessed cumulatively, analysts need to establish why they are being considered together and what effects the combined activity produced. Temporal proximity, shared political narratives and the presence of synthetic media can warrant investigation. A collective-defence assessment requires evidence that supports the much more consequential judgment about the activities and their effects. Russian hybrid threats: Council sanctions twelve individuals and two entities over information manipulation and cyber attacks — Council of the EU — December 2025. Council Decision (CFSP) 2026/2258 — Council of the EU — October 2026. The North Atlantic Treaty, Articles 4–5 — NATO — April 1949. Vilnius Summit Communiqué — NATO — July 2023. Consilium

Cyberterrorism and decision integrity

Analytical schemes for assessing a digital disruption and preserving decision authority during a parallel information incident.

Cyber disruption and the EU terrorism threshold

  1. 1. Establish the actIdentify the intentional conduct and the applicable offence under national law.
  2. 2. Check the specific legal conditionFor system or data interference, apply the relevant cross-reference in Article 3(1)(i). The conditions differ by offence.
  3. 3. Assess gravity and purposeEstablish potential serious harm and evidence of a terrorist purpose under Article 3(2).
  4. 4. Refer for determinationRecord which elements are supported and which remain unresolved. Record any evidenced AI contribution separately.

Disputed information during an operational crisis

Confirm the operational event through an authorised channel, then follow the decision path supported by the evidence.

  • Genuine instruction confirmedReissue the verified instruction through the authorised channel and record the time of confirmation.
  • Authenticity remains unresolvedUse the approved continuity procedure. Preserve the material and obtain independent confirmation before consequential action.
  • Evidence supports higher-level reviewRefer the verified effects, attribution assessment and legal basis to the competent authority.
Evidence and decision table
Four distinct decisions during a combined incident
DecisionEvidence requiredResponsible authority
Protect or restore serviceVerified operational impact, affected services and continuity priorities.Service owner and designated incident command.
Issue or correct an instructionAuthenticated origin, verified underlying facts and a record of publication time.Designated public or operational communications authority.
Assess a terrorism offenceEvidence addressing the offence, applicable legal conditions and specified purpose.Competent investigative, prosecutorial and judicial authorities under national law.
Consider international escalationCorroborated effects, attribution with stated confidence and an identified legal or policy basis.Competent national, EU or Allied authorities, according to the instrument.

Reference documents — EU Counter-Terrorism Directive, Article 3, EU Directive on attacks against information systems, Articles 4, 5 and 9, Fourth EEAS FIMI threat report, North Atlantic Treaty, Articles 4 and 5.

The decisive test is whether interference changed the decisions available in time

An after-action review should reconstruct three timelines in parallel. The first records the underlying operational incident and verified service effects. The second records the appearance, genuine circulation and authentication of disputed information. The third records when each authority received evidence, made a decision and issued an instruction. Aligning those timelines permits a more defensible inference about effect. If responders already faced an outage of their ordinary communications, some delay may have occurred regardless of the information operation. If a verified record shows that an apparently genuine instruction caused an authorised team to suspend a protective step or wait for confirmation beyond a consequential deadline, the case for decision impact becomes stronger. Time to authenticate, time to communicate an authoritative instruction and documented decision delay are therefore the principal measures for a crisis review. Exposure and engagement data can help explain how material reached the people involved. Public attribution and international escalation should retain their own evidentiary thresholds, even when an operational effect has been demonstrated. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — March 2026. Vilnius Summit Communiqué — NATO — July 2023. REPORT ON FIMI THREATS March 2026

Key judgments. The evidence supports confidence that AI can increase the volume and adaptability of manipulative material. Evidence of audience persuasion or operational delay is more limited and must be established for the specific incident. During a crisis, authenticated communications and a clear record of decision authority provide a practical defence even while responsibility remains unresolved.

What would change the assessment. A verified chronology showing that synthetic or manipulated information delayed a warning, interrupted service restoration or changed an authorised decision would provide stronger evidence of cognitive effect than additional publication or engagement counts. Repeated independent cases would justify treating decision delay as a recurring operational risk.

Open official record. Public monitoring establishes coordinated activity, infrastructure and selected engagement observations. The outstanding evidence is a time-stamped incident review connecting a disputed message to an actual decision, identifying what was known at the time and distinguishing its effect from operational delays that were already present. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — March 2026. Rokh Solis technical report — VIGINUM/SGDSN — June 2026. REPORT ON FIMI THREATS March 2026


Pillar III — European Implementation and Choices Through 2031

Chapter 5. Italy, France, Germany, the United Kingdom and EU–NATO Coordination

Europe’s most credible route to military cyber readiness is to make national systems work together while each country retains authority over its missions, data and operating decisions. France’s military programming law incorporates classified AI compute. Germany’s Open Defense Cloud programme is designed around infrastructure that can be deployed with a formation. Italy’s planning documents address the boundary between military cyber responsibilities and the national cyber crisis system. The United Kingdom has created a funded delivery taskforce and published results from separate government cyber-defence pilots. The practical opportunity is to agree on the minimum evidence and permissions that can cross those boundaries. A receiving organisation should be able to determine what was verified, who owns the action, whether it may reuse the information and when a decision is due. That is achievable before all four countries agree on an AI supplier or cloud architecture. The 17 November 2026 endorsement of a European secure defence cloud as one of eight collaborative opportunities gives this work a political opening. The endorsement establishes a planning opportunity. Funding, participating authorities, delivery milestones and operational acceptance still require their own records. Loi n° 2026-791 actualisant la programmation militaire pour les années 2024 à 2030 — Légifrance — August 2026. BWI realisiert Open Defense Cloud für die Bundeswehr — BWI — September 2026. Atto di Indirizzo 2026–2028 — Italian Ministry of Defence — 2025 edition. Rapid AI Delivery Taskforce — UK Ministry of Defence — August 2026. 2026 Defence Review: EU Ministers endorse eight collaborative opportunities — European Defence Agency — November 2026. Légifrance

Italy — The civil–military handover is the first test

Italy’s Atto di Indirizzo 2026–2028 sets out a path to update the military cyber framework and calls for Defence to coordinate with other national actors from peacetime. It envisages the exercise of cyber crisis-management responsibilities for the defence of the state jointly with the National Cybersecurity Agency, or ACN, in the framework of Legislative Decree No. 138/2024. A separate institutional development concerns AI oversight. Legislative Decree No. 179 of 7 October 2026, in force from 23 October, concerns the powers of national authorities under the EU AI Act. Those records make it important to distinguish the authority that assesses a system’s regulatory conformity, the authority that coordinates a cyber crisis and the authority that can direct a military operational action. These functions can intersect in the same incident. A civilian provider may hold the records needed to explain a service failure affecting Defence, while only the relevant military authority can assess the operational consequences. A joint exercise would be most revealing if it measured what information actually crossed that boundary, under which authority, and how long it took for a named receiving officer to accept ownership. Atto di Indirizzo 2026–2028 — Italian Ministry of Defence — 2025 edition. Decreto legislativo 4 settembre 2024, n. 138 — Normattiva — October 2024 publication. Decreto legislativo 7 ottobre 2026, n. 179 — Gazzetta Ufficiale — October 2026. AI_FINAL CON FIRMA

Italy’s industrial and fiscal records point to the decisions that follow. Leonardo presented its Cyber Defence Suite at Eurosatory on 15 June 2026 and identified potential applications across land, naval, aviation and space systems. That is evidence of a supplier offering, with acceptance, deployment and field performance to be established separately. On 11 November 2026, the Chamber of Deputies’ Defence Committee recorded €38,823.1 million in proposed final expenditure appropriations for the Ministry of Defence for 2027, in budget-competence terms, under the draft 2027–2029 budget. That figure describes a proposed ministry-wide appropriation. It provides no programme allocation for military AI or cyber defence. Italy’s concrete opportunity is to use one jointly specified service-continuity problem to determine which part of the proposed solution the state can accept and sustain. A work package with a named civilian service owner, military authority, maintenance commitment and acceptance record would provide a stronger basis for a recurrent budget than a cross-domain product demonstration on its own. Eurosatory, Leonardo presenta la Cyber Defence Suite — Leonardo — June 2026. Bollettino delle Giunte e delle Commissioni parlamentari, Commissione IV, p. 70 — Camera dei deputati — November 2026. leonardo.com

France — Classified compute gives Paris a distinct option for sensitive work

France’s updated military programming law, Loi n° 2026-791 of 16 August 2026, sets out an additional €36 billion rearmament trajectory for 2026–2030. The law’s annex identifies ASGARD as a classified supercomputer dedicated to AI and includes its renewal in the programming trajectory. The €36 billion is a multi-year defence-wide increment in the law’s programme, with no basis for reading it as an AI appropriation or expenditure already incurred. The strategically relevant distinction is between having an environment for sensitive national workloads and being able to deliver an assessed result that an ally can act on. France can pursue the first through national infrastructure. The second depends on who can receive the output, how the finding was validated and whether it remains intelligible when the underlying classified material cannot travel with it. A practical French contribution to a federated European architecture would be to demonstrate that an accredited national environment can produce an authorised, reviewable assessment that a partner can use within its own decision process. The capability to run sensitive workloads domestically strengthens the French position in that negotiation. Its operational value depends on validated workflows, usable capacity and the conditions under which results can be released. Loi n° 2026-791 du 16 août 2026 actualisant la programmation militaire, annex — Légifrance — August 2026. Légifrance

Germany — Deployable infrastructure makes degraded operation a measurable requirement

In September 2026, BWI reported that it was building the Open Defense Cloud for the Bundeswehr on an open-source software basis, with a design intended to be deployable and capable of processing classified data. BWI said applications required for the initial digital capability of a brigade command post were scheduled for 2027, with full operational readiness to be pursued subsequently. The 2027 date is therefore a supplier-reported milestone for an initial application set. The programme gives Germany a direct way to establish what a formation can still do when access to a central service deteriorates. The test would need to specify which data remain available locally, who can authorise their use and how records are reconciled when connectivity returns. Germany’s broader fiscal position provides context but leaves the programme’s lifetime cost unresolved. The Bundestag budget committee record identified €82.69 billion in the regular defence budget and €25.51 billion from the Bundeswehr special fund in the federal government’s 2026 plans. Those are budget figures for two funding channels. A recurring commitment to maintain deployable infrastructure requires its own cost, staffing and readiness evidence. On the civilian side, the BSI dates the entry into force of the German NIS2 implementation law to 6 December 2025 and provides a national registration and reporting channel. An exercise can test the interface with a military contingency while preserving the separate authorities of the two systems. BWI realisiert Open Defense Cloud für die Bundeswehr — BWI — September 2026. Etat 2026: Änderungen im Verteidigungshaushalt beschlossen — Deutscher Bundestag — 2025. Infos zu NIS-2 — Bundesamt für Sicherheit in der Informationstechnik — implementation notice. bwi.de

United Kingdom — Delivery funding and pilot results measure different stages

The UK Ministry of Defence announced the Rapid AI Delivery Taskforce, or TF RAID, in June 2026 and stated in August that the Defence Investment Plan had allocated it £100 million. The remit covers adoption and operational deployment across the military commands, including AI, autonomy and other frontier technologies. A separate 12 October 2026 government announcement said two pilots had helped identify and fix more than 1,000 potential cyber vulnerabilities across government systems. The June case study describes departments validating and prioritising findings through existing processes. The public aggregate leaves the distribution of severity, validation time, closure time and comparative effort unresolved. These two announcements provide different evidence. The taskforce allocation establishes delivery capacity that can be directed towards military priorities. The pilots document a reported result in government environments. For military adoption, the test is which gains transfer to restricted or intermittently connected environments, with their own approval and maintenance constraints. Publishing a severity breakdown and the time to independently verify closure would make the pilots more useful to allies evaluating comparable deployments. Rapid AI Delivery Taskforce — UK Ministry of Defence — August 2026. Britain steps up national defences against growing AI risks — UK Government — October 2026. When AI Leaves the Lab: Testing Frontier Models in Government Cyber Defence — UK Government — June 2026. GOV.UK

The UK also has a distinct route into European coordination. The EU–UK Security and Defence Partnership provides for cooperation on cyber and foreign information manipulation, including exchanges on threat analysis, reporting and responses. The second summit was scheduled for 20 November 2026, two days after this assessment’s cut-off. The operative evidence at the cut-off was the existing partnership and the announced summit agenda. For cyber readiness, a useful implementation measure would be a joint exercise in which the UK and EU-side authorities can exchange verified effects and requests for civilian assistance, while retaining their own decisions on access and escalation. The partnership text supports such cooperation. Participation in a particular EU programme or response mechanism would still depend on that instrument’s terms and the relevant arrangements. Security and Defence Partnership between the European Union and the United Kingdom — May 2025. Prime Minister to attend UK–EU summit in Brussels on 20 November — UK Prime Minister’s Office — October 2026. data.consilium.europa.eu

EU–NATO coordination has working channels and a concrete transport test

The eleventh EU–NATO progress report, covering June 2025 to May 2026, records cyber cooperation under the Structured Dialogue on Cyber, contacts for coordination in a cyber crisis, cooperation between the EU Cyber Defence Coordination Centre and NATO’s Integrated Cyber Defence Centre, and continued exchanges between CERT-EU and the NATO Cyber Security Centre. Those mechanisms have different users and functions. CERT-EU’s technical relationship with its NATO counterpart is one route for exchange. The EU Cybersecurity Reserve provides for support upon eligible request during significant incidents and recovery. Military operational authority remains governed by the applicable national and Alliance arrangements. The test is whether those systems can coordinate when the same incident affects, for example, a civilian transport operator, a movement required by several countries and the public agencies responsible for keeping the transport network operational. Cyber Europe 2026, held on 10–11 June, supplies a concrete exercise setting. ENISA reported more than 5,000 participants and focused the exercise on disruption across interconnected rail and maritime networks. The meaningful next comparison is how long it takes a finding to reach the party empowered to act and whether the civilian and military users receive consistent instructions while the service is degraded. Eleventh progress report on the implementation of the common set of EU–NATO proposals, Cyber security and defence — EU and NATO — June 2026. EU Cybersecurity Reserve — ENISA — programme record. Cyber Europe 2026: All eyes on the EU’s collective response and resilience — ENISA — June 2026. nato.int

JurisdictionDistinct contributionEvidence status as of 18 November 2026Most revealing next test
ItalyCoordination between civilian cyber crisis management and defence responsibilities, with a domestic supplier offering for multiple platform types.Ministry planning direction, AI authorities decree in force, industrial announcement and draft 2027 ministry appropriations.A transport or service incident in which the civilian owner and Defence exchange appropriately authorised findings and close a tracked action. Italian Ministry of Defence planning direction. Chamber Defence Committee budget record. AI_FINAL CON FIRMA
FranceNational infrastructure for classified AI work and a programming basis for its renewal.Enacted military programming update. The annex identifies ASGARD.Release an authorised assessment from a classified national workflow and have a partner verify and use it within its own authority. French military programming law and annex. Légifrance
GermanyAn open-source-based, deployable cloud designed for the Bundeswehr.BWI project announcement and 2027 initial application milestone.A representative command element continues priority work with impaired connectivity, then reconciles records after reconnection. BWI Open Defense Cloud programme. bwi.de
United KingdomA funded military delivery mechanism and separate evidence from government cyber pilots.£100 million taskforce allocation. Government-reported pilot outcomes in a different operating context.Show which pilot benefits transfer to an approved military workflow and publish comparable time, severity and closure measures. UK Rapid AI Delivery Taskforce. UK government cyber-defence pilots. GOV.UK
European Union and NATOCivilian incident assistance, technical exchanges, multinational exercises and military coordination through distinct channels.Documented cooperation and exercise activity. The European secure defence cloud has been endorsed as a collaborative opportunity.A cross-border exercise records which authority received the case, when it accepted responsibility and whether the priority service remained usable. EU–NATO progress report. 2026 EDA Defence Review. nato.int

Key judgments. The four countries are developing different pieces of a usable European capability. The immediate interoperability task is to make verified results, service status and acceptance of responsibility intelligible across those systems. France’s classified infrastructure, Germany’s deployable design, Italy’s crisis-management interface and the United Kingdom’s delivery programmes provide distinct settings in which to test that task.

What would change the assessment. The strongest evidence of convergence would be a jointly accepted record from a real incident or repeated exercise showing that two national authorities and the relevant EU or NATO counterparts completed their part of the handover within the required time, including under degraded conditions.

Open official record. The November CARD endorsement creates an identifiable programme-level question. The next consequential records are a roadmap naming participating states, an agreed accreditation and release framework, a funding decision, and results from a representative test. 2026 Defence Review: EU Ministers endorse eight collaborative opportunities — European Defence Agency — November 2026. eda.europa.eu

Chapter 6. Five-Year Outlook, Indicators, Courses of Action and Net Assessment

Through November 2031, the central outlook is uneven defensive improvement. Some institutions can convert better models into verified remediation, faster restoration and more reliable communications. Others will see the same model capabilities increase the number of cases their existing staff and suppliers must handle. The divergence is likely where access improves faster than authorisation, maintenance and cross-border coordination. The fiscal setting creates room to act, but the connection must be demonstrated at programme level. EDA reports that defence spending across the EU-27 represented 2.2% of GDP in 2025 and projects 2.4% in 2026. Defence investment represented about 32% of defence spending in 2025 and is projected to reach 36% in 2026. These shares describe overall spending and its composition. They identify fiscal space and competing demands, with no breakdown of resources available for defensive AI. The same November review describes a continuing gap between stronger national planning and collective delivery. For 2031, the point is that buying compute and services is only one investment. Maintaining permissions, trained reviewers, deployable communications and tested procedures determines how much of that investment remains useful during disruption. 2026 Coordinated Annual Review on Defence Report, introduction and cooperation findings — European Defence Agency — November 2026. eda.europa.eu

The pathways depend on whether gains survive shared failures

Pathway through November 2031MechanismEvidence that would strengthen itDecision if the evidence emerges
Interoperable defensive gainsNational systems exchange verified outcomes and requests for assistance under agreed release rules. Priority workflows continue through a tested loss of service.Repeated cross-border exercises show shorter handovers, timely restoration and complete decision records under impaired connectivity.Expand the workflows that met acceptance criteria. Fund shared assurance and exercise capacity alongside access to models.
Uneven national improvementBetter-resourced operators reduce remediation time while partners and suppliers continue to face approval, maintenance or staffing delays.Gains concentrate in a few organisations. Comparable records show persistent backlogs or missed handovers at their dependencies.Direct funding and technical assistance towards the bottleneck. Require evidence that the service owner can close the agreed class of cases.
Correlated failure during a compound crisisA common infrastructure dependency becomes unavailable while disputed information delays confirmation of the actual service condition.Alternatives fail in the same exercise, and crisis logs record late authentication or contradictory instructions reaching authorised operators.Use the continuity procedure for the affected function and require proof that the failure path has been addressed before increasing reliance.

These pathways can coexist. One authority can show a durable improvement in remediation while a dependent operator experiences persistent delay. A separate crisis can expose a common dependency across organisations that normally appear independent. The outlook therefore assigns no numerical probabilities. The available evidence supports conditional pathways with testable signs, while the public record supplies neither a comparable set of repeated failures nor a defensible base rate for their occurrence through 2031. The policy value lies in identifying which condition would change an investment decision, rather than assigning a percentage to a compound risk before the relevant systems have been tested. 2026 Coordinated Annual Review on Defence Report — European Defence Agency — November 2026. Eleventh EU–NATO progress report, Cyber security and defence — June 2026. eda.europa.eu

Three decisions can be made before the long-term outcome is known

The first is to establish a joint acceptance standard for a small set of defensive tasks. An exercise design should specify the same evidence and service target for each participating authority, including the point at which the receiving official can accept responsibility for the case. The second is to fund continuity across vulnerable dependencies. For a mission relying on rail, maritime, communications or other civilian services, the responsible authorities should choose the dependency whose failure would most constrain the response and test it directly. The third is to establish a joint decision chronology for compound incidents. Service operators, cyber authorities and authorised communicators need a common record of when the underlying event was verified, when a disputed instruction was authenticated and when the accountable authority issued a decision. The proposed chronology does not require pooling all underlying evidence. It requires reference points that each actor is authorised to share and that can be matched in an after-action review. Together, these decisions support an auditable investment rule. Widen a workflow when it produces verified improvement in the receiving organisation as well as the originating one. Repair the handover, approval or continuity dependency when the gain disappears at the boundary. Cyber Europe 2026: All eyes on the EU’s collective response and resilience — ENISA — June 2026. 2026 Defence Review: EU Ministers endorse eight collaborative opportunities — European Defence Agency — November 2026. ENISA

Course of action and responsible authorityExpected effect and time to effectImplementation burden and reversibilityPrincipal risk and second-order consequence
Joint acceptance exercises. National mission owners lead. NATO or EU counterparts participate within their mandates.An initial assessment is possible in the next agreed exercise cycle. Repeated cycles show whether evidence and responsibility pass reliably between organisations.Moderate. Requires comparable cases and authorised reviewers. Participation can expand incrementally.Scenarios can become too predictable. Rotate dependencies and participating operators so exercise results remain relevant to procurement.
Targeted continuity investment. National service owners and contracting authorities fund the highest-consequence dependencies.Benefits follow successful testing of restoration or service continuity under a defined failure.Higher. May require local infrastructure, supplier obligations and recurrent personnel. Make commitments in stages where possible.Duplicate capacity can raise cost and dilute staffing. A common engineering or identity layer can defeat nominal redundancy.
Shared crisis decision chronology. National crisis leads, designated communicators and eligible EU or NATO counterparts agree what they can record and release.Can improve review and coordination from the first agreed exercise. Operational value depends on adoption by duty teams.Low to moderate. Requires secure retention, agreed time references and authorised release. The record can be revised after exercises.Excess collection can burden responders and unnecessarily expose sensitive information. Limit the shared record to decision-critical facts and references.

European cyber readiness through 2031

Proposed decision sequence for a cross-border incident affecting a service on which civilian and military users depend.

Cross-border handover

  1. 1. Confirm the effectIdentify the affected service, its owner, the verified impact and the next operational deadline.
  2. 2. Authorise releaseIdentify the information that may be shared, its source and any limits on use or onward disclosure.
  3. 3. Obtain acceptanceUse the applicable coordination route. Record the receiving authority, its acknowledgement and who will act.
  4. 4. Reconcile the outcomeRecord continuity or restoration, action taken, unresolved issues and the final decision time.

Decision after the test

  • ExpandUse this path when the receiving authority can act within the mission window and repeat testing confirms the improvement.
  • Repair and retestUse this path when a release rule, approval step or unclear responsibility prevents a timely handover.
  • Limit relianceUse this path when the fallback fails, the service misses its recovery limit or the decision cannot be reconstructed.
Proposed review milestones
Indicators and decision criteria. The dates are proposed review points.
ReviewEvidence to collectDecision supported
2027Matched baseline cases, named service owners and time from report to acknowledged handover.Approve the scope and acceptance criteria for a limited cross-border trial.
2029Repeated handovers during degraded service, recovery time and proof that a tested alternative survives the named failure.Fund expansion where the continuity threshold has been met. Repair and retest failed dependencies.
2031Verified closure, restoration and decision records across multiple operators, including any effect of disputed information.Retain funding for demonstrated operational benefit and prioritise remaining gaps.

Reference documents — Eleventh EU–NATO progress report, EDA 2026 Coordinated Annual Review on Defence, ENISA Cyber Europe 2026.

Indicators that should govern expansion, retention or withdrawal

IndicatorMeasurement and comparisonDecision threshold
Time to accepted cross-border handoverFrom the first verified case record to acceptance by the receiving authority. Compare similar severity, release conditions and cases where an applicable counterpart exists.Expand the workflow when the receiving authority repeatedly accepts the case within the agreed operational window. Review whether tracking excluded difficult cases.
Time from validated finding to verified closureFrom validation to confirmed remediation or an accepted equivalent mitigation. Keep pending cases visible and compare like severity and service constraints.Retain the workflow when closure improves and aged high-priority cases and adverse service effects remain within agreed tolerances.
Continuity after loss of a named dependencyFrom loss of the selected service or infrastructure dependency to resumption of the priority function with required permissions and evidence.Rely on the alternative when it restores the function within the service owner’s stated limit and preserves the decision record.
Delay attributable to disputed informationAlign the incident, authentication and decision timelines. Identify pre-existing service delays and establish which authority relied on which information.Elevate the risk in readiness planning when corroborated cases show that interference altered or delayed a time-critical decision.
Material AI contribution to terrorist cyber disruptionSeek a competent-authority record connecting AI use to the execution or consequences of the cyber conduct, alongside evidence on the relevant offence and purpose.Recalibrate the threat assessment when verified cases demonstrate consequential contribution. Assess recurrence before extrapolating across actors or sectors. European Union Terrorism Situation and Trend Report 2026 — Europol. www.europol.europa.eu

Key judgments. The strongest opportunity is to reduce the interval between verified evidence and authorised action across national and civilian–military boundaries. The most consequential weakness is a dependency shared by a primary system and its nominal alternative. The same reporting standard should track improvements in ordinary workloads and the failures that appear only when connectivity, external services or trusted information are impaired.

What would change the assessment. Repeated evidence that multiple countries can sustain a priority workflow under the same defined disruption would strengthen the more favourable pathway. A series of correlated failures, or a pattern of verified recommendations that remain unactioned at organisational boundaries, would shift the assessment towards uneven adoption and support redirecting resources to the affected dependencies.

Open official record. Four records would materially improve the outlook: the delivery and accreditation roadmap for European secure defence cloud cooperation, independently comparable national acceptance results, a cross-border continuity exercise report with measured handover and recovery times, and incident reviews connecting documented information interference to specific decisions. These would address different uncertainties and should be collected separately. 2026 Defence Review: EU Ministers endorse eight collaborative opportunities — European Defence Agency — November 2026. Cyber Europe 2026 — ENISA — June 2026. eda.europa.eu

Final net assessment

By November 2031, advanced AI should make a material contribution to Allied cyber defence where institutions use it to shorten verified remediation, maintain service continuity and exchange findings that partners can act on. The magnitude of that contribution remains conditional on the capacity of operators and authorities to complete the work once a model has produced an assessment. In terrorism, the established contribution is strongest in media production and adaptation. The claim that AI is increasing serious terrorist cyber disruption requires case-level evidence of conduct, purpose and consequence. In cognitive warfare, the most significant operational risk is that disputed information consumes time needed for an authorised response during an already serious event. For Europe, the decision that connects all three domains is to fund the ability to verify evidence, assign authority and continue operating under disruption. A federation that can preserve those functions across nationally controlled systems will gain more from successive improvements in models and will be better placed to absorb a provider or communications failure. The strongest proof of readiness will be an accepted cross-border decision, protective action completed within its required window, and a record that shows exactly how it happened. 2026 Coordinated Annual Review on Defence Report — European Defence Agency — November 2026. European Union Terrorism Situation and Trend Report 2026 — Europol — July 2026. 4th EEAS Report on Foreign Information Manipulation and Interference Threats — EEAS — March 2026. eda.europa.eu


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