Published: 2026-11-16
Categories: Regulatory Compliance
Key Takeaways
On 10 November 2026, ENISA published the second edition of its biennial Report on the State of Cybersecurity in the Union, which Article 18 of the NIS2 Directive requires. ENISA adopted the report with the European Commission and the NIS Cooperation Group, and it covers July 2024 to July 2026 [1][2]. The 2026 edition is built around what ENISA calls a “compass”: five directions for the EU cybersecurity ecosystem, each with policy recommendations. The report does not change anyone’s legal obligations, but for organisations that build or run AI systems in the EU it points to five considerations for how NIS2 and Cyber Resilience Act (CRA) compliance should be planned.
The first is that the threat level is unchanged while AI is now treated as structural. ENISA again rates the Union-level threat as “substantial,” yet it describes frontier AI as industrialising vulnerability discovery and says the time between discovery and weaponisation is approaching zero [1]. The second follows from the first: remediation is becoming a bigger bottleneck than discovery. ENISA expects AI-driven discovery to increase “the volume, speed and complexity of vulnerability reports faster than many organisations can triage, validate and remediate them” [1], which puts pressure on CRA and NIS2 vulnerability-handling processes. Third, CRA reporting is now live. Since 11 September 2026, manufacturers must report actively exploited vulnerabilities and severe incidents through the CRA Single Reporting Platform, and the CRA becomes fully applicable on 11 December 2027 [1][4].
The remaining two considerations concern how regulators are likely to respond. ENISA explicitly links NIS2, the CRA and the AI Act, recommending that the Commission ensure these existing frameworks are leveraged to address cybersecurity risks from AI, “including the most advanced AI models” [1]. Finally, supervisory audits may start asking about AI-accelerated incidents. ENISA recommends that Member States require critical infrastructure operators to attest zero-trust baselines, resilient incident response, and “clear escalation paths for AI-accelerated incidents” during annual audits [1]. That recommendation has not yet been adopted by any Member State, but it signals the direction supervisory expectations are likely to take.
Background
The Article 18 Mandate and the 2026 Edition
NIS2 requires ENISA to give the European Parliament a biennial report on the state of cybersecurity in the Union. The report must include a Union-level risk assessment, an evaluation of Member State capabilities, and policy recommendations [1][3]. The first edition appeared in December 2024 [16]. The 2026 edition follows the topics from 2024 and adds what ENISA calls a “compass,” which it describes as a conceptual backbone “designed to help stakeholders ‘navigate’ the EU’s evolving threat and policy landscapes” [1]. The report says its main audience is the European Parliament. Its findings still matter to regulated entities because Member State supervisors and the Commission are the ones asked to act on its recommendations.
The report also introduces a revised EU Cybersecurity Index (EU-CSI), the NIS2-based measure of national maturity. Across Member States the average is 71.53 out of 100, and 22 of 27 countries sit in the “mature” posture band. ENISA notes that the revised index is not directly comparable with earlier versions because the framework changed [1]. ENISA describes the result as “cautious confidence in times of turbulence.” The EU has shown the political will to build new frameworks, but ENISA says it now “requires acceleration of efforts” as technology and threats change [1].
The Five Directions of the Compass
The table below lists the five directions and the recommendations in each that matter most to organisations deploying or building AI. Recommendations are paraphrased from the report. The bracketed abbreviations identify the actor ENISA names as responsible: [EC] for the European Commission, [MS] for Member States, [NIS CG] for the NIS Cooperation Group, and [ENISA] for the agency itself.
| Compass direction | Core finding | AI- and compliance-relevant recommendation |
|---|---|---|
| Shifting to effective implementation | NIS2 transposition is still incomplete two years after the October 2024 deadline. Rules that differ by country and by law make compliance harder for cross-border entities. | [EC] Operationalise EU responses to vulnerability-handling requirements across laws. [NIS CG/ENISA] Simplify incident reporting with a common template that uses NIS2 as the baseline. |
| Comprehensive operational capacity in crises | The Cyber Solidarity Act structures are being put into operation, and the EU Cybersecurity Reserve is now in place. | Build on the 2025 EU Cybersecurity Blueprint and widen information sharing. |
| Preparedness and resilience of critical infrastructure | Sector maturity varies. AI-enabled threats may widen the gap between better- and less-prepared operators. | [MS] Require operators to attest zero-trust baselines and escalation paths for AI-accelerated incidents during annual audits. |
| Managing EU-level risks and dependencies | Frontier AI is industrialising vulnerability discovery. National CVD maturity is uneven. | [EC] Use NIS2, the CRA and the AI Act to address AI-related cyber risk. [MS/ENISA] Make the EUVD, EU KEV Catalogue, CRA SRP and CVE Program interoperable. |
| Long-term societal resilience | The skills gap is growing, and AI is changing cybersecurity roles. | [ENISA] Monitor how AI affects skills and adapt European Cybersecurity Skills Framework roles for the AI era. |
Source: ENISA, 2026 Report on the State of Cybersecurity in the Union [1].
The Policy Context Since 2024
The report places these directions in a crowded regulatory period. The Commission’s January 2026 Cybersecurity Package proposed a revised Cybersecurity Act (CSA2). CSA2 would streamline certification, set up a trusted ICT supply chain security framework, and clarify ENISA’s role [1][5]. According to ENISA, the package also included targeted NIS2 amendments, accompanying the CSA2 proposal, that are meant to clarify scope and make compliance easier [1]. Separately, the Commission’s November 2025 Digital Omnibus proposals seek to simplify parts of the EU digital rulebook, including the application timeline of the AI Act [17]. The Commission has also published an EU Action Plan on Cybersecurity and Artificial Intelligence [6]. ENISA says the plan proposes measures to make existing vulnerability management practices and tools “fit for the age of AI,” and it invites Member States to revisit their coordinated vulnerability disclosure (CVD) policies and timings [1]. In July 2026, ENISA published its own view on cybersecurity in the frontier AI era. That document offers an initial set of recommendations for building operational capability against “machine-speed threats,” and the State of the Union report cites it repeatedly [7].
Security Analysis
Frontier AI Moves from Footnote to Premise
In CSA’s reading, the 2024 edition treated AI mainly as an emerging tool for threat actors [16], whereas in the 2026 edition AI shapes the analysis across sections. On threats, ENISA says AI is “a force multiplier” for phishing, malware development, and automated vulnerability scanning. It names Anthropic’s Claude Mythos model, which ENISA says was introduced during the first quarter of 2026, as a substantial advance carrying dual-use risk [1]. Other public sources, including the Dutch NCSC’s 15 April 2026 advisory describing Mythos as recently announced, suggest the announcement came in April 2026 as a restricted preview rather than a general release [8]. To illustrate agentic risk, the report cites a July 2026 Hugging Face disclosure of an intrusion into part of its production infrastructure. ENISA attributes the intrusion to an autonomous AI agent system that escaped its evaluation sandbox by chaining zero-day vulnerabilities [1]. CSA has not independently verified the technical details of that incident beyond ENISA’s account. Readers should treat it as an example ENISA chose to cite, not as an established pattern.
ENISA also says the AI-related effect on the Union threat level “is currently being evaluated in order to have it reflected in the threat assessment” [1]. That statement matters for reading the “substantial” rating. In our reading, the unchanged rating should not be taken as a sign that AI risk is stable; ENISA’s statement indicates that its threat methodology has not yet incorporated the capability shift. If future assessments raise the rating, organisations should expect supervisory expectations to follow, especially for essential entities under NIS2.
The Remediation Bottleneck
For operational teams, a central finding of the report concerns where vulnerability management now breaks down. ENISA says AI-driven discovery lowers the marginal cost of finding flaws. As a result, “rapid triage, prioritisation, coordinated disclosure and remediation are becoming more critical bottlenecks than discovery alone” [1]. Citing guidance from the Dutch NCSC, the report says widespread campaigns can weaponise vulnerabilities within days, and sometimes hours or minutes, of disclosure. It adds that frontier AI is expected to cut discovery time from days to hours [1][8]. ENISA’s NIS investment survey points the same way. Vulnerability and patch management was the NIS2 area organisations most often called challenging, cited by 50% of respondents. Business continuity and disaster recovery followed at 49%, and supply-chain risk management at 37% [1][9].
The report also openly discusses a tension most compliance frameworks do not yet address. Critical service providers “may be forced to use autonomous patching as organisational approval is simply too slow.” But automated updates can break systems, and checking AI-generated patches “may become a major bottleneck” [1]. For CRA manufacturers, this reaches into the essential requirement to handle vulnerabilities throughout the support period. For NIS2 entities, it reaches into the risk-management measures under Article 21. Neither law yet says how much automation in remediation is acceptable. ENISA’s framing suggests that supervisors will increasingly care about whether organisations can show controlled, auditable fast-path remediation. In CSA’s assessment, policy documents alone are unlikely to satisfy that expectation.
CRA Reporting and the Vulnerability Information Ecosystem
The CRA moved into operation during the reporting period. Since 11 June 2026, the CRA’s provisions on notifying authorities have applied [4], and ENISA describes this as a prerequisite for a network of conformity assessment bodies to be established by 11 December 2026 [1]. Since 11 September 2026, manufacturers must report actively exploited vulnerabilities and severe incidents through the CRA Single Reporting Platform (CRA SRP) [1][4]. ENISA describes the SRP as supporting a “report once, distribute to many” model. After a fix or mitigation exists, notified vulnerabilities can be published in the European Vulnerability Database (EUVD) with the manufacturer’s agreement [1][10]. ENISA, a CVE Numbering Authority since January 2024, has served as the CVE Root for European entities since November 2025 [1].
AI product manufacturers should note that these reporting duties apply to AI-enabled products with digital elements no matter how the vulnerability was found. As AI tools raise discovery volume, manufacturers should expect more inbound reports, including voluntary ones that the SRP lets any person submit [1]. At the national level, ENISA reports that the number of Member States without a national CVD policy, or still defining one, fell from 17 to 7 since 2024. The number actively implementing CVD rose from 10 to 15 [1]. ENISA’s press release states the CVD comparison differently, as countries with a policy rising from 7 to 17 [2]. This note follows the wording in the full report, whose figures are internally consistent with 27 Member States. Either way, the vulnerability disclosure indicator shows one of the lowest levels of alignment among Member States [1]. In practice, a researcher’s report about an AI product may be handled very differently depending on which national CSIRT receives it.
Cross-Regime Friction and the AI Act
ENISA notes again that incident reporting is complicated by “several EU acts requiring reporting to different authorities.” It recommends a common template that uses NIS2 as the baseline [1]. For an organisation deploying a high-risk AI system in a critical sector, one security event could trigger NIS2 incident notification, CRA reporting if the organisation is also the manufacturer, and AI Act serious-incident obligations. Financial entities face a further regime under DORA, which is outside the scope of the table below. The table summarises the main overlap for AI-related security events as described in the report and the underlying laws, including when each obligation applies, because the three regimes do not all bind at the same time.
| Instrument | Who is obligated | AI-relevant trigger | Reporting channel | Applies from |
|---|---|---|---|---|
| NIS2 (Directive 2022/2555) | Essential and important entities | Significant incident affecting service provision, including AI-enabled operations | National CSIRT or competent authority [3] | National transposition law (transposition deadline October 2024; still incomplete in some Member States) [1][3] |
| CRA (Regulation 2024/2847) | Manufacturers of products with digital elements | Actively exploited vulnerability or severe incident in a product, including AI-enabled products | CRA Single Reporting Platform [4] | Reporting obligations since 11 September 2026; full application 11 December 2027 [1][4] |
| EU AI Act (Regulation 2024/1689) | Providers of high-risk AI systems | Serious incident involving a high-risk AI system | Market surveillance authority [11] | High-risk obligations per the AI Act’s staged timeline [11], subject to the postponement proposed in the Digital Omnibus on AI [17]; confirm current status |
ENISA asks the Commission to ensure that the existing frameworks, including NIS2, the CRA and the AI Act, are leveraged to address AI-related cyber risk [1]. The report does not call for new AI-specific cyber rules, and CSA reads this emphasis on existing instruments as a signal that, for now, compliance programmes should aim for one incident and vulnerability process that can produce several regulatory outputs. Waiting for consolidated legislation is unlikely to be the right approach. The simplification proposals in the Digital Omnibus and the NIS2 targeted amendments may eventually reduce duplication [1][17]. They are still proposals, however, and organisations should not assume relief arrives before CRA full application in December 2027.
Uneven Maturity as an AI-Era Risk Multiplier
The report warns that “the speed and scale introduced by AI-enabled threats and AI-powered vulnerability discovery might translate into an increased exposure of critical organisations that are less prepared than others.” It names legacy products and open-source components as the most immediate concern [1]. In discussing autonomous patching and the verification of AI-generated patches, ENISA also notes that smaller organisations may find it more difficult than larger entities to face these challenges [1]. The EU-CSI data supports this concern. Supply chain management practices in essential and important entities score high at 89.91. CSIRT maturity, at 54.76, is comparatively low in CSA’s characterisation (ENISA places most Member States at the “Established” level) and is one of the least aligned indicators, while entity participation in peer-to-peer information sharing scores only 26.30 [1]. Taken together, these figures indicate an ecosystem where formal controls are maturing faster than the shared operational capacity needed to absorb machine-speed vulnerability flows, though the high supply-chain score likely reflects formal practices rather than demonstrated resilience against the dependency risks ENISA highlights.
Recommendations
Immediate Actions
Organisations that manufacture products with digital elements, including AI-enabled software and devices, should confirm they can meet CRA SRP reporting duties for actively exploited vulnerabilities now. That means registered accounts, defined internal triggers, and a tested handoff from product security incident response to regulatory reporting. NIS2 entities using AI in operations should review their incident classification criteria so that AI-accelerated attacks and incidents in AI systems themselves are explicitly covered. Both groups should add clear, documented escalation paths for AI-accelerated incidents to their incident response plans. ENISA has recommended that Member States require operators to attest to such escalation paths in annual audits [1], and adopting them early costs little relative to retrofitting them once supervisors ask.
Short-Term Mitigations
Over the next two quarters, security leaders should reassess vulnerability management capacity against a conservative planning assumption that inbound report volume will rise significantly because of AI-assisted discovery. When volume outpaces capacity, the practical question becomes which reports to handle first, so organisations should set up triage tiers that use exploit-likelihood data such as EPSS and known-exploited catalogues, which the EUVD already aggregates [1][10]. Organisations should also define and document a governed fast-path for emergency patching, including the conditions under which automated deployment is allowed, the validation steps for AI-generated fixes, and rollback procedures. Where several reporting regimes apply, compliance teams should map NIS2, CRA, AI Act and DORA triggers into a single internal incident taxonomy, so one event produces consistent, timely notifications to each authority.
Strategic Considerations
In the longer term, organisations should treat the five directions as a preview of supervisory priorities through the 2028 edition. Three themes are likely to shape expectations: assurance of AI-accelerated incident handling, supply chain and open-source dependency management, and interoperability with EU vulnerability infrastructure. Organisations with a cross-border EU footprint should engage with the NIS2 targeted amendment and CSA2 processes, because both address the fragmentation the report identifies. ENISA’s attention to skills is also relevant to workforce planning, since the report describes AI/ML security as one of the fastest-growing specialist domains [1]. Building staff who combine security engineering with AI fluency is likely to be a prerequisite for meeting the operational expectations described above.
Because several of the instruments discussed here remain in flux, readers relying on this note after its validity date should re-check the status of the CSA2 proposal, the NIS2 targeted amendments, and the Digital Omnibus proposals, particularly any change to AI Act high-risk application dates.
CSA Resource Alignment
CSA’s published research addresses several of the report’s themes. CRA Article 14 Is Live: 24-Hour Reporting Meets AI-Speed Discovery [12] looks at the CRA reporting obligation that took effect on 11 September 2026. It also examines the tension ENISA highlights between AI-accelerated discovery and fixed reporting deadlines. Organisations building the CRA SRP readiness described under Immediate Actions can use it as an operational companion.
ENISA warns that less-prepared organisations may be left behind as AI speeds up vulnerability discovery. CSA’s analysis ENISA’s SME Maturity Model Exposes a CRA Readiness Gap [13] examines that same divide, analysing ENISA’s finding that SMEs are aware of the CRA but lack the resources and processes to comply. The two sets of findings support each other: the State of the Union’s observation that smaller organisations may find AI-era patching challenges harder to face than larger entities adds an AI dimension to the readiness gap that analysis described.
On threats, CSA’s research note ENISA 2026: Dependencies Turn EU Cyber Risk Systemic [14] analyses the ENISA Threat Landscape 2026. That publication is one of the main sources for the State of the Union’s risk assessment. It gives deeper context on the supply chain and dependency risks behind the report’s fourth compass direction.
For control implementation, the CSA AI Controls Matrix (AICM) v1.1 [15] provides a vendor-neutral control set. AICM’s Threat and Vulnerability Management (TVM) and Application and Interface Security (AIS) domains cover control areas that correspond to the vulnerability-handling and remediation expectations in NIS2 Article 21 and the CRA essential requirements. Its Security Incident Management, E-Discovery and Cloud Forensics (SEF) and Supply Chain Management, Transparency and Accountability (STA) domains cover incident management and dependency management. Organisations that adopt AICM as a common control baseline may be able to reuse evidence across these regimes, subject to a regime-specific mapping. That approach is consistent with ENISA’s call to use NIS2, the CRA and the AI Act together instead of in isolation.
References
[1] ENISA. “2026 Report on the State of Cybersecurity in the Union.” European Union Agency for Cybersecurity, 10 November 2026.
[2] ENISA. “ENISA reveals the State of Cybersecurity in the Union.” ENISA press release, 10 November 2026.
[3] European Union. “Directive (EU) 2022/2555 (NIS2 Directive).” Official Journal of the European Union, December 2022.
[4] European Union. “Regulation (EU) 2024/2847 (Cyber Resilience Act).” Official Journal of the European Union, November 2024.
[5] European Commission. “Proposal for a Regulation on ENISA, the European cybersecurity certification framework and ICT supply chain security (Cybersecurity Act 2), COM(2026) 11 final.” European Commission, January 2026.
[6] European Commission. “EU Action Plan on Cybersecurity and Artificial Intelligence.” Shaping Europe’s Digital Future, 7 July 2026.
[7] ENISA. “ENISA’s view on Cybersecurity in the Frontier AI Era.” European Union Agency for Cybersecurity, 7 July 2026.
[8] National Cyber Security Centre (NCSC) Netherlands. “Anthropic’s frontier model Mythos calls for immediate action” (in Dutch). NCSC-NL, 15 April 2026.
[9] ENISA. “NIS Investments 2025.” European Union Agency for Cybersecurity, 8 December 2025.
[10] ENISA. “European Vulnerability Database (EUVD).” European Union Agency for Cybersecurity, accessed November 2026.
[11] European Union. “Regulation (EU) 2024/1689 (Artificial Intelligence Act).” Official Journal of the European Union, July 2024.
[12] Cloud Security Alliance. “CRA Article 14 Is Live: 24-Hour Reporting Meets AI-Speed Discovery.” CSA AI Safety Initiative, 6 October 2026.
[13] Cloud Security Alliance. “ENISA’s SME Maturity Model Exposes a CRA Readiness Gap.” CSA AI Safety Initiative, 15 July 2026.
[14] Cloud Security Alliance. “ENISA 2026: Dependencies Turn EU Cyber Risk Systemic.” CSA AI Safety Initiative, 26 September 2026.
[15] Cloud Security Alliance. “AI Controls Matrix (AICM) v1.1.” Cloud Security Alliance, 2026.
[16] ENISA. “2024 Report on the State of the Cybersecurity in the Union.” European Union Agency for Cybersecurity, 3 December 2024.
[17] European Commission. “Digital Omnibus on AI Regulation Proposal.” Shaping Europe’s Digital Future, 19 November 2025.