Regulation explainerCyber Defense
AI cybersecurity regulation in the EU: Article 15, the Cyber Resilience Act and DORA
Three EU instruments set cybersecurity expectations for AI systems. Article 15 of the AI Act requires high-risk systems to be accurate, resilient and secure over their lifecycle. The Cyber Resilience Act covers products with digital elements and already requires vulnerability reporting. DORA covers financial entities and their ICT providers. This page maps who is affected, key dates checked this November and the evidence to prepare.
On this page
- Three instruments, three different questions
- Where each instrument reaches AI systems
- Key dates for AI cybersecurity duties, checked in November
- Status caveats before you plan around these dates
- Which instrument to check first
- Evidence to have ready for a supervisor or customer
- Where the sources say less than people assume
- Questions and answers
- Sources
Three instruments, three different questions
No single EU law is titled AI security. Instead, three instruments each ask a different question. The AI Act asks whether a high-risk AI system is accurate, resilient and secure. The Cyber Resilience Act asks whether a product with digital elements is built and maintained securely. The Digital Operational Resilience Act, known as DORA, asks whether a financial entity can withstand and recover from disruption to its ICT systems.
An organization can sit under more than one at once. A bank that deploys a credit-scoring model is a financial entity under DORA and a deployer of a high-risk system under the AI Act. A software company that ships an AI-enabled product may face the AI Act and the Cyber Resilience Act together. Start by listing which hats you wear, then read the matching row below.
This page covers cybersecurity duties only. For deployer duties such as human oversight and logging, see the guide on EU AI Act deployer obligations.
Where each instrument reaches AI systems
EU AI Act (Regulation (EU) 2024/1689), Article 15
EU; providers of high-risk AI systemsApplies whenThe system is classified as high-risk under the Act. Article 15 sets accuracy, robustness and cybersecurity requirements for such systems1.
- Achieve an appropriate level of accuracy, robustness and cybersecurity, and perform consistently in those respects throughout the lifecycle1.
- Make the system resilient against attempts by unauthorized third parties to alter its use, outputs or performance by exploiting vulnerabilities1.
- Include technical solutions for AI-specific vulnerabilities, including data poisoning, model poisoning, adversarial examples or model evasion, confidentiality attacks and model flaws1.
- Declare accuracy levels and the relevant metrics in the instructions of use1.
Cyber Resilience Act (Regulation (EU) 2024/2847)
EU; manufacturers, importers and distributors of products with digital elementsApplies whenYou place a product with digital elements on the EU market. Whether a given AI-enabled product is in scope is a classification question to settle with counsel23.
- Report actively exploited vulnerabilities and severe incidents under Article 14, which applies from 11 September 2026: an early warning within 24 hours, a notification within 72 hours and a final report within 14 days after a fix is available for exploited vulnerabilities, or within one month for severe incidents2.
- Submit reports through the Single Reporting Platform run by ENISA, which routes them to the relevant CSIRT2.
- Meet the Annex I requirements on product properties and vulnerability handling, including due diligence on third-party components, from 11 December 202723.
DORA, Regulation (EU) 2022/2554
EU; financial entities and their critical ICT third-party providersApplies whenYou are one of the more than 20 types of financial entity in scope4. DORA entered into force on 16 January 2023 and has applied since 17 January 20254.
- Run ICT risk management, third-party risk management, resilience testing including threat-led penetration testing, and incident reporting4.
- Per the European supervisory authorities, treat frontier AI as a source of ICT risk: keep inventories that include AI and ML components, monitor continuously and test and rehearse recovery5.
- Keep the management body responsible for governance and for reviewing the risk appetite framework, with measures proportionate to the entity5.
Key dates for AI cybersecurity duties, checked in November
| Milestone | Date | What it means for AI security |
|---|---|---|
| AI Act enters into force | 1 August 20246 | Starting point for all later AI Act dates. |
| AI Omnibus in force | 27 July 20266 | Moved the high-risk dates below; the Commission page records the proposal of 19 November 2025 and political agreement on 7 May 20266. |
| High-risk rules under Annex III | 2 December 20276 | Article 15 applies to systems in the Annex III use cases from this date. |
| High-risk rules for products under Annex I | 2 August 20286 | Article 15 applies to AI embedded in regulated products from this date. |
| CRA vulnerability and incident reporting | 11 September 20262 | Already applying: the early warning clock for actively exploited vulnerabilities runs now2. |
| CRA main obligations | 11 December 20272 | Annex I product requirements apply to products placed on the market from this date. |
| DORA application | 17 January 20254 | Already applying to financial entities. |
Dates come from the Commission and ESMA pages listed in the sources. Confirm the current status before relying on any date.
Status caveats before you plan around these dates
Which instrument to check first
- If
You provide an AI system used in an Annex III area such as hiring or credit decisions.
- If
You sell software or connected hardware in the EU that includes an AI model.
ThenAsk counsel whether the product is a product with digital elements and build the Article 14 reporting route immediately.
Reporting duties have applied since 11 September 20262.
- If
You are a financial entity adopting AI assistants or agents.
ThenAdd AI and ML components to your ICT asset inventory and test scenarios, and brief the management body.
The supervisory authorities say DORA and the AI Act provide a foundation and that firms should act fast and proactively5.
- If
You only use a third-party AI service internally and ship nothing.
ThenCheck your sector rules and supplier contracts first, and treat Article 15 as a benchmark when you assess the provider.
Article 15 is addressed to providers of high-risk systems rather than to every user of AI1.
Evidence to have ready for a supervisor or customer
These items map one security record to several instruments. Each should exist before a regulator or buyer asks.
Where the sources say less than people assume
Two limits are worth stating plainly. First, the ESAs' statement of 31 July 2026 says its annex does not establish additional requirements; it encourages financial entities to apply existing DORA and AI Act duties with urgency5. Treat it as supervisory direction, not new law.
Second, Article 15 does not prescribe a particular test or tool. It sets outcomes: appropriate robustness and cybersecurity, resilience to specified attacks and declared accuracy1. How you evidence those outcomes is your design choice, and a repeatable red team and a documented threat model are sensible places to start.
Questions and answers
Does Article 15 of the EU AI Act apply to every AI system?
No. Article 15 sets accuracy, robustness and cybersecurity requirements for high-risk AI systems1. Whether your system is high-risk depends on its classification under the Act. Systems outside that category are not subject to Article 15 directly, although customers and sector regulators may still expect similar security practices.
What does Article 15 say about attacks on AI models?
It requires technical solutions for AI-specific vulnerabilities, naming data poisoning, model poisoning, adversarial examples or model evasion, confidentiality attacks and model flaws1. It also requires resilience against unauthorized third parties who try to alter a system's use, outputs or performance by exploiting vulnerabilities1.
When do the AI Act high-risk rules apply after the AI Omnibus?
The Commission's page lists 2 December 2027 for Annex III high-risk rules and 2 August 2028 for Annex I product-embedded systems, via the AI Omnibus6. The Omnibus entered into force on 27 July 20266. Confirm the current dates before planning, because supervisory guidance and national measures may add detail.
Do Cyber Resilience Act reporting duties apply to AI products now?
Article 14 reporting has applied since 11 September 2026 to manufacturers within the Act's scope: an early warning within 24 hours, then a notification within 72 hours2. Whether a specific AI-enabled product counts as a product with digital elements needs a scoping decision with counsel.
How does DORA relate to AI use in banks and insurers?
Sources
- Article 15: Accuracy, robustness and cybersecurity — European Commission, AI Act Service Desk · checked 19 November 2026
- Safer and more secure digital products — European Commission · checked 19 November 2026
- Cyber Resilience Act summary — European Commission · checked 19 November 2026
- Digital Operational Resilience Act (DORA) — ESMA · checked 19 November 2026
- Toward a consistent and risk-based approach for ICT risks from frontier AI models (JC 2026 25) — European Supervisory Authorities · checked 19 November 2026
- AI Act: regulatory framework for AI — European Commission · checked 19 November 2026
More in Cyber Defense
Back to Cyber DefenseYour starting point
What would you like to move forward?
Start with the outcome that feels closest. We'll help give the next step a useful shape.
Let's find your next move
One conversation. A clearer direction.
Tell us what you want to improve. A few sentences are enough to start.
shayan@coldai.org