Side-by-side
| Dimension | AI Act | NIS2 |
|---|---|---|
| Scope | AI systems placed on the EU market or used in the EU. | Essential and important entities in critical sectors (Annex I + Annex II of NIS2). |
| Trigger | System falls into Article 5 (prohibited), Article 6(1) (Annex I product), Article 6(2) (Annex III use), or is GPAI. | Entity meets the size and sector thresholds in NIS2 Articles 2-3. |
| Substantive controls | Risk management (Art 9), data governance (Art 10), documentation (Art 11), oversight (Art 14), robustness/cybersecurity (Art 15). | Cybersecurity risk management measures (Art 21), incident handling, supply-chain security, business continuity. |
| Conformity | Annex VI internal or notified-body assessment depending on the use case (Art 43). | Self-declared compliance with Article 21 measures; supervised by the competent authority. |
| Incident reporting | 15-day serious-incident report (Art 73); 2-day for critical-infrastructure cases that breach Union law or affect fundamental rights. | Early warning 24h; notification 72h; final report 1 month (Art 23). |
| Maximum fine | EUR 35M or 7% global turnover (Art 99(1) prohibited); EUR 15M or 3% (Art 99(2) high-risk); EUR 7.5M or 1% (Art 99(3) incorrect information). | EUR 10M or 2% (essential entities); EUR 7M or 1.4% (important entities) under Art 34. |
| Enforcement deadline | Article 50 transparency: 2 August 2026. Annex III high-risk: 2 August 2026 as written, provisionally 2 December 2027 (Digital Omnibus agreement of 7 May 2026, pending formal adoption). | National transposition deadline 17 October 2024; enforcement live in most Member States. |
Where they overlap
- Annex III §2 critical infrastructure AI. The AI Act's substantive controls apply to the AI itself; NIS2 Article 21 applies to the entity that runs the AI. Both regimes address robustness and incident handling but at different layers.
- Cybersecurity controls. Article 15 of the AI Act sets the floor for the AI system's own cybersecurity. NIS2 Article 21(2) sets the floor for the entity's overall security posture, including how it operates the AI. Article 15 evidence does not satisfy NIS2 21; NIS2 21 evidence does not satisfy Article 15.
- Supply chain. NIS2 Article 21(2)(d) requires supply-chain risk management. When the AI vendor is the third party, the AI Act provider's Article 13 instructions and Annex IV documentation feed the NIS2 supply-chain due diligence.
- Incident reporting. Both regimes require it. The clocks differ. The information differs (NIS2 focuses on operational impact; AI Act on the AI failure mode). Most operators end up running parallel reports through both channels.
Where they conflict (in practice)
There is no doctrinal conflict — the regulations address different addressees and different obligations. The friction is operational:
- The 24-hour NIS2 early warning is materially shorter than the AI Act's 15-day serious-incident window. Operators must triage during the first 24 hours whether the event is reportable under NIS2 only, AI Act only, or both — and start the NIS2 clock immediately.
- NIS2 competent authorities (CSIRTs, BSI, ANSSI) are not yet AI Act market surveillance authorities in most Member States. Cross-authority coordination is still being built.
- The Article 21(2) NIS2 measures and Article 15 robustness measures must both be evidenced in the technical file but use different vocabularies and audit expectations.
Practical compliance for an entity in both
- Maintain a unified incident-response runbook keyed to the 24-hour NIS2 clock, with an AI Act overlay for any incident touching the AI system.
- Cross-reference your NIS2 Article 21 self-assessment with the AI Act provider's Article 13 instructions for use; document any gaps.
- Include AI providers in the NIS2 supply-chain register and the DORA register where applicable.
- For dual-supervised entities, request a joint inspection plan from the NIS2 competent authority and the AI Act market surveillance authority — both supervisors have the legal authority to inspect the same system.
Which regulation takes precedence
Neither. Both apply simultaneously. Where they address the same outcome from different angles (cybersecurity, incident reporting), the stricter requirement controls in practice. The Recital 64 of the AI Act and Recital 90 of NIS2 confirm the regimes are complementary.
Cross-regulatory data update
Auto-merged from the Fontvera archetype dataset on 2026-05-12. The sections below are extracted verbatim from the `obligations` and `obligation_crossrefs` tables; the page itself was last reviewed manually before this update.
Summary statistics
Overlaps: 3 · Conflicts: 0 · Gaps: 2
5 article-level crossrefs catalogued between AI Act and NIS2 Directive from the Fontvera EU regulatory corpus. Article numbers are verbatim from the underlying obligation_crossrefs table; descriptions are extracted, not paraphrased.
All crossrefs between these regulations
| Article (A) | Article (B) | Type | Severity | Description |
|---|---|---|---|---|
| AI Act Art 15 | NIS2 Directive Art 21 | overlap | high | [entity affected: Provider of high-risk AI systems / Essential or important entity] Both regulations require entities to implement technical and organizational measures to ensure cybersecurity, robust |
| AI Act Art 12 | NIS2 Directive Art 21 | overlap | medium | [entity affected: Provider of high-risk AI systems / Essential or important entity] Both regulations mandate logging and monitoring capabilities to record events, facilitate incident response, and mai |
| AI Act Art 11 | NIS2 Directive Art 21 | overlap | medium | [entity affected: Provider of high-risk AI systems / Essential or important entity] Both regulations require the maintenance of technical documentation and records to demonstrate compliance with secur |
| AI Act Art ? | NIS2 Directive Art ? | gap | high | [entity affected: AI System Integrators in Critical Infrastructure] Entities integrating AI into critical infrastructure may face undefined liability if the AI provider is non-EU and the integrator is |
| AI Act Art ? | NIS2 Directive Art ? | gap | low | [entity affected: Small AI Startups in Non-Critical Sectors] Small AI providers operating outside high-risk categories and non-critical sectors may lack clear guidance on baseline cybersecurity hygien |
Overlaps explained
No conflict-type crossrefs were catalogued for this pair, but the 3 overlaps below mean a single control can be designed to satisfy both regulations at once. Plan the controls jointly to avoid duplicate effort:
- AI Act Art 15 vs NIS2 Directive Art 21 (high severity) — [entity affected: Provider of high-risk AI systems / Essential or important entity] Both regulations require entities to implement technical and organizational measures to ensure cybersecurity, robustness, and resilience of their systems against errors, faults, and cyber threats.
- AI Act Art 12 vs NIS2 Directive Art 21 (medium severity) — [entity affected: Provider of high-risk AI systems / Essential or important entity] Both regulations mandate logging and monitoring capabilities to record events, facilitate incident response, and maintain situational awareness regarding system operations and security.
- AI Act Art 11 vs NIS2 Directive Art 21 (medium severity) — [entity affected: Provider of high-risk AI systems / Essential or important entity] Both regulations require the maintenance of technical documentation and records to demonstrate compliance with security and risk management requirements for regulatory authorities.
Which regulation takes precedence
EU law does not lay down a universal precedence rule between AI Act and NIS2 Directive. In practice three resolution approaches apply: lex specialis (the more specific provision wins when both purport to govern the same conduct); regulator guidance (EDPB, EBA, ESMA and the AI Office have all issued joint readings on overlapping articles — check the most recent applicable opinion); and document the choice (when the regulations leave the call to the controller, the audit defence is your written reasoning, not the regulator's silence). Where the corpus surfaces a conflict rather than an overlap, treat that as an escalation path to legal — not a control-design question.