AI Act vs CRA
Where they overlap, where they differ and which asks more, topic by topic, with their deadlines and fines side by side.
At a Glance
They differ on 28 of 34 topics. Each verdict names what it rests on; read the wording before relying on it.
- AI Act Asks more on 0 topics Alone on 12 topics More detailed on 1 topic
- CRA Asks more on 0 topics Alone on 12 topics More detailed on 3 topics
| Topic | AI Act | CRA | Verdict |
|---|---|---|---|
| Governance | |||
| Roles and independence | AI Act: 3 | CRA: 1 | AI Act more detailedmore requirements on this, not necessarily stricter |
| Awareness and training | AI Act: 1 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Risk management | |||
| Risk management framework | AI Act: 1 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Risk assessment | AI Act: 1 | CRA: 1 | Both cover itread the wording below |
| Review, audit and improvement | AI Act: 1 | CRA: 5 | CRA more detailedmore requirements on this, not necessarily stricter |
| Asset management | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Protection | |||
| Access control | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Authentication | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Cryptography and encryption | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Data security | AI Act: 1 | CRA: 2 | Both cover itread the wording below |
| Secure acquisition and development | AI Act: 1 | CRA: 2 | Both cover itread the wording below |
| Configuration, change and patch management | AI Act: – | CRA: 2 | Only CRAno specific requirement on the other side in the acts covered |
| Vulnerability handling | AI Act: – | CRA: 7 | Only CRAno specific requirement on the other side in the acts covered |
| Detection and response | |||
| Monitoring and detection | AI Act: 2 | CRA: 1 | Both cover itread the wording below |
| Incident handling | AI Act: 2 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Incident classification | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Reporting to authorities | AI Act: 4 | CRA: 4 | Both cover itread the wording below |
| Informing customers and the public | AI Act: – | CRA: 4 | Only CRAno specific requirement on the other side in the acts covered |
| Threat intelligence and information sharing | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Continuity and resilience | |||
| Business continuity and crisis management | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Security testing | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Threat-led penetration testing | AI Act: 1 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Third parties and supply chain | |||
| Supplier and third-party risk | AI Act: 2 | CRA: 4 | CRA more detailedmore requirements on this, not necessarily stricter |
| Contract requirements | AI Act: 1 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Privacy and data protection | |||
| Privacy by design and by default | AI Act: – | CRA: 1 | Only CRAno specific requirement on the other side in the acts covered |
| Impact assessment | AI Act: 1 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Artificial intelligence | |||
| Risk classification of AI systems | AI Act: 4 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Data and data governance | AI Act: 3 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Transparency to users | AI Act: 6 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Human oversight | AI Act: 2 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Conformity assessment and monitoring | AI Act: 5 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Supervision and enforcement | |||
| Registration with authorities | AI Act: 2 | CRA: – | Only AI Actno specific requirement on the other side in the acts covered |
| Cooperation with supervisors | AI Act: 1 | CRA: 7 | CRA more detailedmore requirements on this, not necessarily stricter |
| Penalties and liability | AI Act: 4 | CRA: 3 | Both cover itread the wording below |
What This Covers
The obligations below are summarised from the acts listed here, grouped by topic. Where a side shows nothing on a topic, the acts covered have no specific requirement on it; an act not yet included, national guidance or a supervisor's expectations may still say more. Use it to see where the regimes differ, not as a compliance checklist.
Included
- AI Act
- KI-loven Draft act on artificial intelligence (KI-loven), consultation of 30 June 2025
- Decreet 16 oktober 2026 Decreet van 16 oktober 2026, Belgisch Staatsblad 19 November 2026, numac 2026008360
- No. 31/2026 Regulation of Artificial Intelligence Act 2026 (No. 31 of 2026)
Status
Reviewed against the official texts.
Included
- CRA Regulation (EU) 2024/2847 of the European Parliament and of the Council of 23 October 2024
- CRA Product Categories Commission Implementing Regulation (EU) 2025/2392 of 28 November 2025 on the technical description of the categories of important and critical products with digital elements pursuant to Regulation (EU) 2024/2847 of the European Parliament and of the Council (OJ L, 2025/2392, 1.12.2025)
- Kamerstuk 36875 Kamerstukken 36875, Regels ter uitvoering van Verordening (EU) 2024/2847 (Uitvoeringswet verordening cyberweerbaarheid), voorstel van wet (36875, nr. 2) and memorie van toelichting (36875, nr. 3)
- S.L. 591.06 L.N. 238 of 2026, Cyber Resilience Regulations, 2026, made under article 58 of the Malta Digital Innovation Authority Act (Cap. 591); Government Gazette No. 21,717, 11 September 2026; S.L. 591.06
Status
Reviewed against the official texts.
Key Numbers
Deadline
AI Act: Immediately on identifying a serious incident (Deployers of high-risk AI systems); Without undue delay (Providers of general-purpose AI models with systemic risk); Within 2 days of becoming aware (Providers of high-risk AI systems); Within 10 days of becoming aware (Providers of high-risk AI systems)
CRA: Final report within 14 days of a corrective or mitigating measure being available
Threshold
AI Act: Not in the acts covered
CRA: Harm to protection of sensitive or important data or functions, or malicious code
Goes to
AI Act: Market surveillance authorities and, where applicable, the notified body (ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.); Provider first, then importer or distributor and market surveillance authorities (Deployers of high-risk AI systems); Market surveillance authority (Public law bodies and private providers of public services deploying Article 6(2) systems (except Annex III, point 2), and deployers of Annex III, points 5(b) and (c) systems)
CRA: Manufacturer and market surveillance authorities (Importers); Manufacturer and market surveillance authorities (Distributors)
Evidence to keep
AI Act: Documented assessment, provided on request (Providers of Annex III AI systems they consider not high-risk); Annex IV technical documentation (Providers of high-risk AI systems); Automatically generated logs, at least six months (ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.); Documentation, 10 years after placing on the market or putting into service (ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.); Provider contact details, EU declaration, technical documentation and any notified body certificate, 10 years (Providers of high-risk AI systems established in third countries); Certificate, instructions and EU declaration, 10 years (Importers of high-risk AI systems); Logs under the deployer's control, at least six months (Deployers of high-risk AI systems); EU declaration of conformity, 10 years (Providers of high-risk AI systems); Annex XI and Annex XII documentation (Providers of general-purpose AI models); Post-market monitoring plan in the technical documentation (Providers of high-risk AI systems)
CRA: Documented risk assessmentA structured process for identifying risks, analysing their possible effects and judging which need attention. In information security, this means examining threats, weaknesses and existing safeguards to prioritise risks to information and business operations. The assessment informs decisions about reducing or accepting risk and should be reviewed as circumstances change. in the technical documentation; EU declaration of conformity, at least 10 years or the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. if longer (Importers); Technical documentation and EU declaration, at least 10 years or the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. if longer; Software bill of materialsA formal record of the details and supply chain relationships of the components included in the software of a product. Under the CRA, manufacturers must draw up an SBOM in a commonly used, machine-readable format covering at least the top-level dependencies.
Incident reports go to
AI Act: AI Office and, as appropriate, national competent authorities (Providers of general-purpose AI models with systemic risk); Market surveillance authorities of the Member States where the incident occurred (Providers of high-risk AI systems); Market surveillance authorities of the Member States where the incident occurred (Providers of high-risk AI systems); Market surveillance authorities of the Member States where the incident occurred (Providers of high-risk AI systems)
CRA: Coordinating CSIRTComputer Security Incident Response Team and ENISAEuropean Union Agency for Cybersecurity, via the single reporting platform; Coordinating CSIRTComputer Security Incident Response Team and ENISAEuropean Union Agency for Cybersecurity, via the single reporting platform; Coordinating CSIRTComputer Security Incident Response Team of the main establishment and ENISAEuropean Union Agency for Cybersecurity
First notice of an incident
AI Act: Within 2 days of becoming aware for a widespread infringement or critical infrastructure disruption; 10 days for a death; 15 days otherwise (Providers of high-risk AI systems)
CRA: Within 24 hours of becoming aware; Within 24 hours of becoming aware Shortest stated
Incident notification
AI Act: Not in the acts covered
CRA: Within 72 hours of becoming aware; Within 72 hours of becoming aware
Intermediate report
AI Act: Not in the acts covered
CRA: On request of the coordinating CSIRTComputer Security Incident Response Team; On request of the coordinating CSIRTComputer Security Incident Response Team
Final report
AI Act: Not in the acts covered
CRA: Within one month of the incident notification
Deadline to have the measures in place
AI Act: By 2 December 2026 (Operators of AI systems); By 2 December 2027 (Annex III) and 2 August 2028 (Annex I) (Providers of AI systems); By 2 December 2027 (Annex III) and 2 August 2028 (Annex I) (ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.); By 2 December 2026 for systems on the market before 2 August 2026 (Providers of AI systems, including general-purpose AI systems, that generate synthetic content)
CRA: Not in the acts covered
External assessment or certification
AI Act: Internal control, or a notified body where Article 43 requires one (Providers of high-risk AI systems)
CRA: Module A, modules B and C, module H, or an EU certification scheme; Third-party assessment or certification, depending on class (Manufacturers of important and critical products)
Review of the risk framework
AI Act: Regular systematic review throughout the lifecycle (Providers of high-risk AI systems)
CRA: Not in the acts covered
Registration with the authority
AI Act: EU database, under Article 49(2) (Providers of Annex III AI systems they consider not high-risk); EU database under Article 71, before placing on the market or putting into service (Providers of Annex III AI systems or their authorised representatives); EU database under Article 71, before putting into service or use (Deployers that are public authorities, Union bodies or persons acting on their behalf)
CRA: Not in the acts covered
Maximum fines
AI Act: Up to EUR 35,000,000 or 7% of worldwide annual turnover, whichever is higher (Operators of AI systems); Up to EUR 15,000,000 or 3% of worldwide annual turnover, whichever is higher (Operators of AI systems and notified bodies); Up to EUR 7,500,000 or 1% of worldwide annual turnover, whichever is higher (Anyone supplying information to notified bodies or national competent authorities in reply to a request); Up to EUR 15,000,000 or 3% of worldwide annual turnover, whichever is higher (Providers of general-purpose AI models) Highest stated
CRA: Up to EUR 15,000,000 or 2.5% of worldwide annual turnover, whichever is higher (Manufacturers); Up to EUR 10,000,000 or 2% of worldwide annual turnover, whichever is higher (Economic operators and others subject to the listed articles); Up to EUR 5,000,000 or 1% of worldwide annual turnover, whichever is higher (Economic operators)
Governance
Roles and independence
AI Act more detailed more requirements on this, not necessarily stricter
NIST CSF 2.0 GV.RR · ISO/IEC 27001:2022 A.5.2, A.5.3
Provider obligations
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Ensure the system meets the Section 2 requirements, taking account of its intended purpose and the state of the art. Show your name and contact address, run a quality management system, keep documentation and logs, complete conformity assessment, the EU declaration and CE marking, register, take corrective action, demonstrate conformity on request and meet accessibility requirements.
Appoint an authorised representative
Applies to: Providers of high-risk AI systems established in third countries
Before making a high-risk system available in the Union, appoint an EU-established authorised representative by written mandate. The representative verifies the conformity documents, keeps them for 10 years, deals with authorities and ends the mandate, informing the authority, if the provider acts against the Regulation.
When others become the provider
Applies to: Distributors, importers, deployers, other third parties and initial providers
A distributor, importer, deployer or other third party takes on provider obligations if it puts its name or trademark on a high-risk system (unless a contract allocates them otherwise), substantially modifies one, or changes the purpose of an AI system so that it becomes high-risk. The initial provider then stops being the provider of that system but cooperates and hands over information and access, unless it clearly specified that the system is not to be changed into a high-risk one.
When others become the manufacturer
Applies to: Importers and distributors who sell a product under their own name or trademark or substantially modify it, and others who substantially modify a product
An importer or distributor that sells a product under its own name or trademark, or substantially modifies it, is treated as the manufacturer and is subject to Articles 13 and 14. Any other person who substantially modifies a product and makes it available is also a manufacturer, for the affected part or the whole product if its overall cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. is affected.
Awareness and training
Only AI Act no specific requirement on the other side in the acts covered
NIST CSF 2.0 PR.AT · ISO/IEC 27001:2022 A.6.3
Support AI literacy
Take measures to support the AI literacy of staff and others who operate or use AI systems on your behalf, taking account of their knowledge, experience and training, the context of use and the people affected. The duty does not require you to guarantee a specific level of literacy for any individual.
Nothing in the acts covered.
Risk management
Risk management framework
Only AI Act no specific requirement on the other side in the acts covered
NIST CSF 2.0 GV.RM · ISO/IEC 27001:2022 Clause 6.1
Risk management system
Applies to: Providers of high-risk AI systems
Establish, implement, document and maintain a risk management system for each high-risk AI system. Run it as a continuous, iterative process across the whole lifecycle, with regular systematic review and updating.
Nothing in the acts covered.
Risk assessment
Both cover it read the wording below
NIST CSF 2.0 ID.RA · ISO/IEC 27001:2022 Clauses 6.1.2, 8.2
Assess, treat and test for risk
Applies to: Providers of high-risk AI systems
Identify and evaluate known and foreseeable risks to health, safety and fundamental rights, including under foreseeable misuse and from post-market data, and adopt targeted measures until residual risk is acceptable. Consider the effect on under-18s and other vulnerable groups. Test against predefined metrics and probabilistic thresholds, in any event before placing the system on the market or putting it into service.
Cybersecurity risk assessment
Assess the product's cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. risks and use the result across planning, design, development, production, delivery and maintenance. Document it, update it during the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years., base it on intended and foreseeable use, and include it in the technical documentation with a justification for any requirement that does not apply.
Review, audit and improvement
CRA more detailed more requirements on this, not necessarily stricter
NIST CSF 2.0 ID.IM, GV.OV · ISO/IEC 27001:2022 A.5.35, A.5.36, Clauses 9, 10
Quality management system
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Run a documented quality management system with written policies and procedures, covering regulatory strategy, design and development controls, testing, data management, risk management, post-market monitoring, incident reporting, record-keeping, resources and an accountability framework for management and staff. Scale it to the size of the organisation, without lowering the level of protection.
Correct non-conformity
If you know or have reason to believe the product or your processes do not meet Annex I, immediately take corrective measures, or withdraw or recall the product as appropriate.
EU declaration of conformity
Once conformity is shown, draw up the EU declaration, following the Annex V structure, and supply it or a simplified version with the product. Use a single declaration where several Union acts apply. By drawing it up you assume responsibility for the product's compliance.
Technical documentation
Draw up technical documentation with at least the Annex VII elements before placing the product on the market and update it as needed, at least during the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years.. Keep it and the EU declaration available to the market surveillance authorities for at least 10 years after the product is placed on the market, or for the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. if longer.
Choose a conformity route
Assess the product and your processes against Annex I using internal control (module A), EU-type examination with conformity to type (modules B and C), full quality assurance (module H), or a European cybersecurity certification schemeA comprehensive set of requirements, technical requirements, standards and procedures set at EU level for the certification or conformity assessment of specific ICT products, services or processes. Schemes are adopted under the Cybersecurity Act, and NIS2 and the CRA can require their use. where available and applicable.
Third-party assessment by product class
Applies to: Manufacturers of important and critical products
Important products in class I use modules B and C or module H unless harmonised standards, common specifications or a scheme at assurance level substantial were fully applied. Class II use modules B and C, module H or a scheme at that level. Critical products use a certification scheme under Article 8(1), or the class II procedures where its conditions are not met.
Asset management
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 ID.AM · ISO/IEC 27001:2022 A.5.9 to A.5.13
Nothing in the acts covered.
Software bill of materials
Identify and document vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. and components, including a software bill of materialsA formal record of the details and supply chain relationships of the components included in the software of a product. Under the CRA, manufacturers must draw up an SBOM in a commonly used, machine-readable format covering at least the top-level dependencies. in a commonly used, machine-readable format covering at least the top-level dependencies.
Protection
Access control
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 PR.AA · ISO/IEC 27001:2022 A.5.15, A.5.18, A.8.2, A.8.3
Nothing in the acts covered.
Protect against unauthorised access
Where applicable, protect against unauthorised access with appropriate controls such as authenticationThe process of checking that a user, device or software process is who or what it claims to be before allowing access to a system. A system may check evidence such as a password or a security credential. Authentication establishes identity; access requirements then determine what that identity may do. Multi-factor authentication is one approach to authentication, using more than one kind of evidence., identity or access management, and report possible unauthorised access.
Authentication
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 PR.AA · ISO/IEC 27001:2022 A.5.17, A.8.5
Nothing in the acts covered.
Protect against unauthorised access
Where applicable, protect against unauthorised access with appropriate controls such as authenticationThe process of checking that a user, device or software process is who or what it claims to be before allowing access to a system. A system may check evidence such as a password or a security credential. Authentication establishes identity; access requirements then determine what that identity may do. Multi-factor authentication is one approach to authentication, using more than one kind of evidence., identity or access management, and report possible unauthorised access.
Cryptography and encryption
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 PR.DS · ISO/IEC 27001:2022 A.8.24
Nothing in the acts covered.
Confidentiality and integrity of data
Where applicable, protect the confidentialityThe information-security property of keeping information from being disclosed to people, organisations or systems that are not authorised to receive it. Access restrictions and encryption help preserve confidentiality when information is stored, sent or processed. It concerns who can see information; it does not by itself ensure that the information is accurate or that a service remains available. of stored, transmitted or processed data, such as by state-of-the-art encryptionEncryption changes readable information into an unreadable form using a mathematical method and a key, a value that controls the transformation. Decryption recovers the original information with the appropriate key. Encryption helps keep stored or transmitted data confidential, so people without the necessary key cannot readily read it. at rest and in transit. Protect data, commands, programs and configuration against unauthorised change, and report corruption.
Data security
Both cover it read the wording below
NIST CSF 2.0 PR.DS · ISO/IEC 27001:2022 A.5.12, A.8.10 to A.8.12
Protect the model and its infrastructure
Applies to: Providers of general-purpose AI models with systemic risk
Ensure an adequate level of cybersecurity protection for the model and its physical infrastructure.
Confidentiality and integrity of data
Where applicable, protect the confidentialityThe information-security property of keeping information from being disclosed to people, organisations or systems that are not authorised to receive it. Access restrictions and encryption help preserve confidentiality when information is stored, sent or processed. It concerns who can see information; it does not by itself ensure that the information is accurate or that a service remains available. of stored, transmitted or processed data, such as by state-of-the-art encryptionEncryption changes readable information into an unreadable form using a mathematical method and a key, a value that controls the transformation. Decryption recovers the original information with the appropriate key. Encryption helps keep stored or transmitted data confidential, so people without the necessary key cannot readily read it. at rest and in transit. Protect data, commands, programs and configuration against unauthorised change, and report corruption.
Minimise data and allow removal
Where applicable, process only data that is adequate, relevant and limited to the product's purpose. Let users securely and easily remove all data and settings permanently, and make any transfer to other products secure.
Secure acquisition and development
Both cover it read the wording below
NIST CSF 2.0 PR.PS · ISO/IEC 27001:2022 A.8.25 to A.8.31
Accuracy, robustness and cybersecurity
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Achieve an appropriate level of accuracy, robustness and cybersecurity throughout the lifecycle and declare accuracy metrics in the instructions. Make the system resilient to errors and to attempts to alter its use or output, with measures where appropriate against data poisoning, model poisoning, adversarial examples, confidentiality attacks and model flaws.
Meet the essential requirements
When placing a product on the market, ensure it was designed, developed and produced in line with the cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. requirements in Annex I, Part I.
No known vulnerabilities, secure by default and reduced attack surface
Where applicable on the basis of the risk assessmentA structured process for identifying risks, analysing their possible effects and judging which need attention. In information security, this means examining threats, weaknesses and existing safeguards to prioritise risks to information and business operations. The assessment informs decisions about reducing or accepting risk and should be reviewed as circumstances change., supply the product without known exploitable vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. and with a secure-by-default configuration that can be reset. Limit attack surfaces and reduce the impact of incidents with exploitation mitigation.
Configuration, change and patch management
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 PR.PS · ISO/IEC 27001:2022 A.8.9, A.8.19, A.8.32
Nothing in the acts covered.
Keep security updates available
Keep each security update issued during the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. available for at least 10 years after it is issued, or for the rest of the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. if longer.
Test, fix and disclose vulnerabilities
Remediate vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without delay, with security updates separate from functional ones where feasible, distributed securely and free of charge unless agreed otherwise for tailor-made products. Test and review security regularly. Enforce a coordinated vulnerability disclosureA process in which someone who discovers a security weakness reports it to the system owner or product supplier and works with them to address it. The parties coordinate investigation, a fix and the timing and content of any public disclosure, reducing the chance of exposing users before protection is available. A national cybersecurity centre may help coordinate when several parties are involved or contact fails. policy, provide a reporting contact, and publish information on fixed vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition..
Vulnerability handling
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 ID.RA · ISO/IEC 27001:2022 A.8.8
Nothing in the acts covered.
Third-party components
Exercise due diligence when integrating third-party components, including open-source ones, so they do not compromise the product. If you find a vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. in a component, report it to whoever makes or maintains the component and remediate it, sharing your fix where you developed one.
Set and state the support period
Handle vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. effectively for a support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. that reflects expected use and is at least five years, or the expected use time if shorter. State the end date, at least month and year, clearly at the time of purchase, and notify users when support ends where technically feasible.
Report actively exploited vulnerabilities
Notify any actively exploited vulnerabilityA vulnerability for which there is reliable evidence that a malicious actor has exploited it in a system without the owner's permission. Under the CRA, manufacturers must send an early warning within 24 hours of becoming aware of one in their product, through the single reporting platform. you become aware of to the coordinating CSIRTComputer Security Incident Response Team and ENISAEuropean Union Agency for Cybersecurity at the same time. Send an early warning within 24 hours and a vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. notification with general information and mitigations within 72 hours. Send a final report describing the vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition., any malicious actor and the fix no later than 14 days after a corrective or mitigating measure is available. The coordinating CSIRTComputer Security Incident Response Team may ask for an intermediate report.
Importer corrective action and alerts
Applies to: Importers
Hold back non-conforming products and take corrective measures, withdraw or recall those already placed. Inform the manufacturer of any vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without undue delay, and where a product presents a significant cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. risk, immediately inform the market surveillance authorities.
Distributor corrective action and alerts
Applies to: Distributors
Make sure corrective measures, withdrawal or recall take place for non-conforming products. Inform the manufacturer of any vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without undue delay. Immediately inform market surveillance authorities of a significant cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. risk. If the manufacturer has ceased operations, inform the authorities without undue delay and, as far as possible, users.
Software bill of materials
Identify and document vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. and components, including a software bill of materialsA formal record of the details and supply chain relationships of the components included in the software of a product. Under the CRA, manufacturers must draw up an SBOM in a commonly used, machine-readable format covering at least the top-level dependencies. in a commonly used, machine-readable format covering at least the top-level dependencies.
Test, fix and disclose vulnerabilities
Remediate vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without delay, with security updates separate from functional ones where feasible, distributed securely and free of charge unless agreed otherwise for tailor-made products. Test and review security regularly. Enforce a coordinated vulnerability disclosureA process in which someone who discovers a security weakness reports it to the system owner or product supplier and works with them to address it. The parties coordinate investigation, a fix and the timing and content of any public disclosure, reducing the chance of exposing users before protection is available. A national cybersecurity centre may help coordinate when several parties are involved or contact fails. policy, provide a reporting contact, and publish information on fixed vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition..
Detection and response
Monitoring and detection
Both cover it read the wording below
NIST CSF 2.0 DE.CM, DE.AE · ISO/IEC 27001:2022 A.8.15, A.8.16
Automatic logs
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Build the system so it records events automatically over its lifetime, supporting risk identification, post-market monitoring and deployer monitoring. Keep the logs under your control for a period that suits the intended purpose and is at least six months, unless other law says otherwise.
Keep deployer logs
Applies to: Deployers of high-risk AI systems
Keep the logs generated by the system, to the extent you control them, for a period that suits the intended purpose and is at least six months, unless other law says otherwise. Financial institutions keep them with their financial services documentation.
Record and monitor activity
Where applicable, provide security-related information by recording and monitoring relevant internal activity, including access to or changes of data, services or functions, with a user opt-outAn opt-out is a way for a person to refuse an activity or stop further participation. In direct marketing, it means taking a step such as unsubscribing from messages. In a default-on arrangement, the activity proceeds unless the person refuses. Having an opt-out does not establish valid consent where the applicable law requires a positive agreement beforehand..
Incident handling
Only AI Act no specific requirement on the other side in the acts covered
NIST CSF 2.0 RS.MA, RS.AN, RS.MI · ISO/IEC 27001:2022 A.5.24 to A.5.28
Corrective action and duty to inform
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
If a system may not conform, immediately correct, withdraw, disable or recall it and tell distributors and, where applicable, deployers, the authorised representative and importers. If it presents a risk under Article 79(1), immediately investigate the causes and inform the market surveillance authorities and, where applicable, the notified body.
Investigate a serious incident
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
After reporting, investigate without delay, including a risk assessment and corrective action, and cooperate with authorities. Do not alter the system in a way that could affect later evaluation of the causes without first informing the authorities.
Nothing in the acts covered.
Incident classification
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 RS.MA · ISO/IEC 27001:2022 A.5.25
Nothing in the acts covered.
When an incident is severe
Treat an incident as severe if it harms or could harm the product's ability to protect the availabilityThe information-security property that authorised people and systems can access and use information or services when needed, reliably and within an acceptable time. An outage or an attack that overwhelms a service can reduce availability even if no information is stolen or changed. In product security, this includes keeping essential functions accessible and usable., authenticity, integrity or confidentialityThe information-security property of keeping information from being disclosed to people, organisations or systems that are not authorised to receive it. Access restrictions and encryption help preserve confidentiality when information is stored, sent or processed. It concerns who can see information; it does not by itself ensure that the information is accurate or that a service remains available. of sensitive or important data or functions, or if it has introduced or could introduce malicious code into the product or a user's systems.
Reporting to authorities
Both cover it read the wording below
NIST CSF 2.0 RS.CO · ISO/IEC 27001:2022 A.5.5
Report serious incidents on models
Applies to: Providers of general-purpose AI models with systemic risk
Track, document and report relevant information about serious incidents and possible corrective measures without undue delay.
Report a serious incident
Applies to: Providers of high-risk AI systems
Report any serious incident immediately after establishing a causal link, or its reasonable likelihood, between the system and the incident, taking account of its severity. An incomplete initial report may be followed by a complete one.
Report widespread infringements and critical infrastructure disruption
Applies to: Providers of high-risk AI systems
Report immediately in the event of a widespread infringement, or a serious and irreversible disruption of the management or operation of critical infrastructure.
Report a death
Applies to: Providers of high-risk AI systems
Where a person dies, report immediately after establishing, or as soon as you suspect, a causal relationship between the system and the incident.
Report actively exploited vulnerabilities
Notify any actively exploited vulnerabilityA vulnerability for which there is reliable evidence that a malicious actor has exploited it in a system without the owner's permission. Under the CRA, manufacturers must send an early warning within 24 hours of becoming aware of one in their product, through the single reporting platform. you become aware of to the coordinating CSIRTComputer Security Incident Response Team and ENISAEuropean Union Agency for Cybersecurity at the same time. Send an early warning within 24 hours and a vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. notification with general information and mitigations within 72 hours. Send a final report describing the vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition., any malicious actor and the fix no later than 14 days after a corrective or mitigating measure is available. The coordinating CSIRTComputer Security Incident Response Team may ask for an intermediate report.
Report severe incidents
Notify any severe incident affecting the product's security to the coordinating CSIRTComputer Security Incident Response Team and ENISAEuropean Union Agency for Cybersecurity at the same time. Send an early warning within 24 hours stating whether unlawful or malicious acts are suspected, and an incident notification with an initial assessment and mitigations within 72 hours. Send a final report with the root cause and measures within one month of the incident notification. The coordinating CSIRTComputer Security Incident Response Team may ask for an intermediate report.
Reporting route
Submit notifications through ENISAEuropean Union Agency for Cybersecurity's single reporting platform, using the end-point of the coordinating CSIRTComputer Security Incident Response Team in the Member State of your main establishment. Manufacturers without one follow the order in Article 14(7). You may ask for dissemination of a sensitive notification to be delayed on justified cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. grounds.
Voluntary reporting Optional
Applies to: Manufacturers and any other natural or legal person
Any person may voluntarily notify a coordinating CSIRTComputer Security Incident Response Team or ENISAEuropean Union Agency for Cybersecurity of vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition., cyber threatsAny potential circumstance, event or action that could damage, disrupt or otherwise harm network and information systems, their users or other persons. NIS2, DORA, the CRA, the Network Code on Cybersecurity and the Cyber Solidarity Act all use this Cybersecurity Act definition., incidents and near missesAn event that could have compromised the availability, authenticity, integrity or confidentiality of data or of services offered by network and information systems, but was prevented or did not materialise. NIS2 allows voluntary reporting of near misses, and the CRA uses the same definition.. Doing so creates no additional obligations for the person notifying.
Informing customers and the public
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 RS.CO, RC.CO
Nothing in the acts covered.
Set and state the support period
Handle vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. effectively for a support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. that reflects expected use and is at least five years, or the expected use time if shorter. State the end date, at least month and year, clearly at the time of purchase, and notify users when support ends where technically feasible.
Product identity, contact point and instructions
Give the product a type, batch or serial number and show your name and contact details. Designate a single point of contact for users that is not limited to automated tools. Supply the Annex II information and instructions and keep them available for at least 10 years or the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years., whichever is longer.
Cooperate with authorities and announce cessation
On a reasoned request, give the market surveillance authority the information needed to show conformity and cooperate on measures to remove risks. If you are ceasing operations and cannot comply, inform the authorities and, as far as possible, users before this takes effect.
Inform users
After becoming aware of an actively exploited vulnerabilityA vulnerability for which there is reliable evidence that a malicious actor has exploited it in a system without the owner's permission. Under the CRA, manufacturers must send an early warning within 24 hours of becoming aware of one in their product, through the single reporting platform. or severe incident, inform impacted users, and where appropriate all users, and tell them about mitigations they can deploy, where appropriate in a structured, machine-readable format.
Threat intelligence and information sharing
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 ID.RA · ISO/IEC 27001:2022 A.5.6, A.5.7
Nothing in the acts covered.
Voluntary reporting Optional
Applies to: Manufacturers and any other natural or legal person
Any person may voluntarily notify a coordinating CSIRTComputer Security Incident Response Team or ENISAEuropean Union Agency for Cybersecurity of vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition., cyber threatsAny potential circumstance, event or action that could damage, disrupt or otherwise harm network and information systems, their users or other persons. NIS2, DORA, the CRA, the Network Code on Cybersecurity and the Cyber Solidarity Act all use this Cybersecurity Act definition., incidents and near missesAn event that could have compromised the availability, authenticity, integrity or confidentiality of data or of services offered by network and information systems, but was prevented or did not materialise. NIS2 allows voluntary reporting of near misses, and the CRA uses the same definition.. Doing so creates no additional obligations for the person notifying.
Continuity and resilience
Business continuity and crisis management
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 RC.RP, RC.CO · ISO/IEC 27001:2022 A.5.29, A.5.30
Nothing in the acts covered.
Availability of essential functions
Where applicable, protect the availabilityThe information-security property that authorised people and systems can access and use information or services when needed, reliably and within an acceptable time. An outage or an attack that overwhelms a service can reduce availability even if no information is stolen or changed. In product security, this includes keeping essential functions accessible and usable. of essential and basic functions, also after an incident, including resilience against denial-of-service attacks. Minimise the product's negative impact on services provided by other devices or networks.
Security testing
Only CRA no specific requirement on the other side in the acts covered
NIST CSF 2.0 ID.IM · ISO/IEC 27001:2022 A.5.35, A.8.29, A.8.34
Nothing in the acts covered.
Test, fix and disclose vulnerabilities
Remediate vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without delay, with security updates separate from functional ones where feasible, distributed securely and free of charge unless agreed otherwise for tailor-made products. Test and review security regularly. Enforce a coordinated vulnerability disclosureA process in which someone who discovers a security weakness reports it to the system owner or product supplier and works with them to address it. The parties coordinate investigation, a fix and the timing and content of any public disclosure, reducing the chance of exposing users before protection is available. A national cybersecurity centre may help coordinate when several parties are involved or contact fails. policy, provide a reporting contact, and publish information on fixed vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition..
Threat-led penetration testing
Only AI Act no specific requirement on the other side in the acts covered
NIST CSF 2.0 ID.IM
Evaluate models and mitigate systemic risk
Applies to: Providers of general-purpose AI models with systemic risk
Evaluate the model using standardised protocols and tools reflecting the state of the art, including documented adversarial testing. Assess and mitigate possible systemic risks at Union level and their sources.
Nothing in the acts covered.
Third parties and supply chain
Supplier and third-party risk
CRA more detailed more requirements on this, not necessarily stricter
NIST CSF 2.0 GV.SC · ISO/IEC 27001:2022 A.5.19, A.5.21 to A.5.23
Importer checks
Applies to: Importers of high-risk AI systems
Before placing a high-risk system on the market, verify the conformity assessment, technical documentation, CE marking, EU declaration, instructions and authorised representative. Hold back a system you have reason to think is non-conforming or falsified, and report risks. Keep the certificate, instructions and declaration for 10 years and cooperate with authorities.
Distributor checks
Applies to: Distributors of high-risk AI systems
Before making a high-risk system available, verify the CE marking, EU declaration and instructions, that the provider and importer show their name and contact address, and that the provider has a quality management system. Hold back a system you consider non-conforming. If one is already on the market, ensure it is corrected, withdrawn or recalled, inform the provider or importer and authorities of any risk, and cooperate with them.
Third-party components
Exercise due diligence when integrating third-party components, including open-source ones, so they do not compromise the product. If you find a vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. in a component, report it to whoever makes or maintains the component and remediate it, sharing your fix where you developed one.
Importer checks
Applies to: Importers
Place on the market only products and manufacturer processes that meet Annex I. First ensure the conformity assessmentThe process of verifying whether a product, service or system meets the requirements set by the applicable EU act, for example the essential cybersecurity requirements of the CRA or the high-risk requirements of the AI Act. Depending on the risk class, the manufacturer does it alone or with a notified body. and technical documentation exist, the product has CE markingA marking by which a manufacturer or provider indicates that a product conforms to the applicable EU harmonisation legislation. Under the AI Act, high-risk AI systems must bear the CE marking, and the same mechanism applies under other product laws., EU declaration and instructions, and the identification and support-date duties are met. Show your own contact details on the product, its packaging or an accompanying document.
Distributor checks and cooperation
Applies to: Distributors
Act with due care. Before making a product available, verify the CE markingA marking by which a manufacturer or provider indicates that a product conforms to the applicable EU harmonisation legislation. Under the AI Act, high-risk AI systems must bear the CE marking, and the same mechanism applies under other product laws. and that the manufacturer and importer met their identification, instruction, support-date and declaration duties. Hold back non-conforming products and cooperate with the authority on reasoned request.
Software bill of materials
Identify and document vulnerabilitiesA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. and components, including a software bill of materialsA formal record of the details and supply chain relationships of the components included in the software of a product. Under the CRA, manufacturers must draw up an SBOM in a commonly used, machine-readable format covering at least the top-level dependencies. in a commonly used, machine-readable format covering at least the top-level dependencies.
Contract requirements
Only AI Act no specific requirement on the other side in the acts covered
NIST CSF 2.0 GV.SC · ISO/IEC 27001:2022 A.5.20
Written agreements with suppliers
Applies to: Providers of high-risk AI systems and their third-party suppliers
Specify by written agreement the information, capabilities, technical access and other assistance a supplier of AI systems, models, tools, services, components or processes gives so that the provider can comply. This does not apply to open-source tools, services, processes or components that are made public, other than general-purpose AI models.
Nothing in the acts covered.
Privacy and data protection
Privacy by design and by default
Only CRA no specific requirement on the other side in the acts covered
NIST Privacy Framework CT.DP-P
Nothing in the acts covered.
Minimise data and allow removal
Where applicable, process only data that is adequate, relevant and limited to the product's purpose. Let users securely and easily remove all data and settings permanently, and make any transfer to other products secure.
Impact assessment
Only AI Act no specific requirement on the other side in the acts covered
NIST Privacy Framework ID.RA-P
Fundamental rights impact assessment
Applies to: Public law bodies and private providers of public services deploying Article 6(2) systems (except Annex III, point 2), and deployers of Annex III, points 5(b) and (c) systems
Before first use, assess the impact on fundamental rights: the processes involved, period and frequency of use, people affected, specific risks of harm, human oversight, and measures if risks materialise, including governance and complaints. Update it when elements change and notify the market surveillance authority of the results. You may cross-refer to an existing data protection impact assessment.
Nothing in the acts covered.
Artificial intelligence
Risk classification of AI systems
Only AI Act no specific requirement on the other side in the acts covered
Prohibited AI practices
Applies to: Operators of AI systems
Do not place on the market, put into service or use AI that manipulates people or exploits their vulnerabilities causing, or likely to cause, significant harm, scores people socially leading to unjustified or out-of-context detriment, predicts crime from profiling alone, builds facial recognition databases by untargeted scraping, infers emotions at work or in education other than for medical or safety reasons, or categorises people biometrically to infer sensitive traits. Real-time remote biometric identification in public spaces for law enforcement is banned outside narrow exceptions.
Intimate imagery and child sexual abuse material
Applies to: Operators of AI systems
Do not place on the market, put into service or use AI that generates or manipulates realistic non-consensual intimate or sexually explicit material of an identifiable person, or child sexual abuse material unless a national 'without right' defence applies. For providers the ban bites where this is the intended purpose, or a foreseeable and reproducible outcome without adequate safeguards. For deployers it bites where the system is used for that purpose. Manipulation that adds no exposure and does not alter the nature of the activity is outside the ban.
Classify high-risk AI systems
Applies to: Providers of AI systems
Treat an AI system as high-risk where it is a safety component of a product, or is itself a product, under Annex I legislation that requires third-party conformity assessment, or where it falls in an Annex III use case. A third-party assessment required only for risks other than health and safety does not count. AI used only for non-safety assistance, optimisation, efficiency, convenience or quality control is not a safety component, unless its failure would endanger health and safety.
Document an Annex III opt-out
Applies to: Providers of Annex III AI systems they consider not high-risk
An Annex III system is not high-risk only if it poses no significant risk and meets one of four conditions, such as a narrow procedural or preparatory task. It is always high-risk if it profiles people. A provider relying on this documents its assessment before placing the system on the market or putting it into service, registers it and gives the assessment to authorities on request.
Nothing in the acts covered.
Data and data governance
Only AI Act no specific requirement on the other side in the acts covered
Special category data for bias correction Optional
Providers may exceptionally process special categories of personal data where strictly necessary to detect and correct bias. Other data, including synthetic or anonymised data, must not be able to do the job, and the data must be secured, not passed to others, deleted once the bias is corrected or the retention period ends, and the reasons recorded.
Data sets and data governance
Applies to: Providers of high-risk AI systems
Apply governance to training, validation and testing data covering design choices, origin, preparation, data gaps and examination for bias, with measures to detect, prevent and mitigate it. Data sets must be relevant, sufficiently representative, as error-free and complete as possible, and fit the intended setting. Where no model is trained, this applies to testing data only.
Copyright policy and training content summary
Applies to: Providers of general-purpose AI models
Put in place a policy to comply with Union copyright law, including identifying and respecting rights reservations. Publish a sufficiently detailed summary of the content used for training, using the AI Office template.
Nothing in the acts covered.
Transparency to users
Only AI Act no specific requirement on the other side in the acts covered
Transparency and instructions for use
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Design the system so deployers can interpret its output and use it properly. Supply clear instructions covering the provider's identity, intended purpose, tested accuracy, robustness and cybersecurity, foreseeable risks, human oversight measures, maintenance needs and how to collect and interpret logs.
Tell workers and affected people
Applies to: Deployers of high-risk AI systems, including employers
Before using a high-risk system at the workplace, employers inform workers' representatives and affected workers. Deployers of Annex III systems that make or assist decisions about people tell those people that the system is used.
Disclose interaction with AI
Applies to: Providers of AI systems that interact directly with people
Design systems that interact directly with people so those people are told they are dealing with AI, unless this is obvious to a reasonably well-informed person. Give the information clearly at the latest at the first interaction, in an accessible form.
Mark synthetic content
Applies to: Providers of AI systems, including general-purpose AI systems, that generate synthetic content
Ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, with solutions that are effective, interoperable, robust and reliable as far as technically feasible. This does not apply to assistive standard editing or where the input is not substantially altered.
Disclose emotion recognition and deep fakes
Applies to: Deployers of emotion recognition, biometric categorisation or content-generating AI systems
Inform people exposed to emotion recognition or biometric categorisation. Disclose that deep fake content, and text published to inform the public on matters of public interest, is artificially generated or manipulated. Artistic works carry a limited disclosure, and reviewed text under editorial responsibility is exempt.
Model documentation
Applies to: Providers of general-purpose AI models
Keep up-to-date technical documentation of the model for the AI Office and national authorities (Annex XI), and information for providers integrating the model (Annex XII). Open-source models with public parameters are exempt unless they carry systemic risk.
Nothing in the acts covered.
Human oversight
Only AI Act no specific requirement on the other side in the acts covered
Design for human oversight
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Design the system so people can effectively oversee it in use: understand its limits, watch for automation bias, interpret output, disregard or override it, and stop it safely. For remote biometric identification under Annex III, point 1(a), no action is taken on an identification unless at least two competent people verify it separately. This does not apply in law enforcement, migration, border control or asylum where Union or national law considers it disproportionate.
Use as instructed and assign oversight
Applies to: Deployers of high-risk AI systems
Use the system in line with its instructions through appropriate technical and organisational measures. Assign human oversight to people with the necessary competence, training, authority and support. Where you control input data, ensure it is relevant and sufficiently representative for the intended purpose.
Nothing in the acts covered.
Conformity assessment and monitoring
Only AI Act no specific requirement on the other side in the acts covered
Technical documentation
Applies to: Providers of high-risk AI systems
Draw up technical documentation with at least the Annex IV elements before placing the system on the market or putting it into service, and keep it up to date. SMEs and small mid-caps may use the Commission's simplified form. Products under Annex I, Section A legislation need a single set of documentation.
Monitor use and report problems
Applies to: Deployers of high-risk AI systems
Monitor operation on the basis of the instructions. If use may present a risk, inform the provider or distributor and the market surveillance authority without undue delay and suspend use. If you identify a serious incident, immediately inform the provider first, then the importer or distributor and the market surveillance authorities.
Conformity assessment
Applies to: Providers of high-risk AI systems
Annex III, point 1 systems may choose internal control or a notified body where harmonised standards or common specifications are fully applied, and otherwise use a notified body. Annex III, points 2 to 8 use internal control, and Annex I, Section A systems follow the sectoral procedure. A substantial modification needs a new assessment. For systems that continue to learn, changes pre-determined at the initial assessment and documented are not a substantial modification.
EU declaration of conformity and CE marking
Applies to: Providers of high-risk AI systems
Draw up an EU declaration for each system with the Annex V content, keep it up to date and available for 10 years, and accept responsibility for compliance by drawing it up. Affix the CE marking visibly, legibly and indelibly, or digitally for systems provided digitally, with the notified body number where applicable.
Post-market monitoring
Applies to: Providers of high-risk AI systems
Establish and document a post-market monitoring system, proportionate to the technology and risk, that actively collects and analyses performance data throughout the system's lifetime to evaluate continuing compliance. Base it on a plan held in the technical documentation.
Nothing in the acts covered.
Supervision and enforcement
Registration with authorities
Only AI Act no specific requirement on the other side in the acts covered
Register in the EU database
Applies to: Providers of Annex III AI systems or their authorised representatives
Before placing on the market or putting into service an Annex III high-risk system, or a system you concluded is not high-risk under Article 6(3), register yourself and the system in the EU database. Critical infrastructure systems under Annex III, point 2 are registered at national level.
Public bodies register their use
Applies to: Deployers that are public authorities, Union bodies or persons acting on their behalf
Before putting into service or using an Annex III high-risk system, other than a critical infrastructure system under point 2, register yourself, select the system and register its use in the EU database. Do not use a system that is not registered, and inform the provider or distributor.
Nothing in the acts covered.
Cooperation with supervisors
CRA more detailed more requirements on this, not necessarily stricter
Keep documents and cooperate with authorities
Applies to: ProvidersProvider Under the AI Act of high-risk AI systemsAn AI system classified as high-risk under AI Act Article 6, either through the Annex I product-safety route or an Annex III use case. Certain Annex III systems may fall outside that classification under Article 6(3) if its conditions are met, while Annex III systems performing profiling of natural persons remain high-risk. Classification determines the applicable conformity and compliance requirements.
Keep technical and quality system documentation, notified body decisions and the EU declaration available to authorities for 10 years. On a reasoned request, give the authority the information needed to show conformity and access to logs under your control.
Cooperate with authorities and announce cessation
On a reasoned request, give the market surveillance authority the information needed to show conformity and cooperate on measures to remove risks. If you are ceasing operations and cannot comply, inform the authorities and, as far as possible, users before this takes effect.
Importer corrective action and alerts
Applies to: Importers
Hold back non-conforming products and take corrective measures, withdraw or recall those already placed. Inform the manufacturer of any vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without undue delay, and where a product presents a significant cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. risk, immediately inform the market surveillance authorities.
Importer records and cooperation
Applies to: Importers
Keep the EU declaration of conformity for at least 10 years after the product is placed on the market, or for the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. if longer, and ensure the technical documentation can be made available on request. Provide information to the authority on reasoned request and cooperate. If the manufacturer has ceased operations, inform the authorities and, where possible, users.
Distributor checks and cooperation
Applies to: Distributors
Act with due care. Before making a product available, verify the CE markingA marking by which a manufacturer or provider indicates that a product conforms to the applicable EU harmonisation legislation. Under the AI Act, high-risk AI systems must bear the CE marking, and the same mechanism applies under other product laws. and that the manufacturer and importer met their identification, instruction, support-date and declaration duties. Hold back non-conforming products and cooperate with the authority on reasoned request.
Distributor corrective action and alerts
Applies to: Distributors
Make sure corrective measures, withdrawal or recall take place for non-conforming products. Inform the manufacturer of any vulnerabilityA weakness, susceptibility or flaw that a threat can exploit. NIS2 applies it to ICT products and services, the CRA to products with digital elements and DORA to any asset, system, process or control. Coordinated vulnerability disclosure and handling duties build on this definition. without undue delay. Immediately inform market surveillance authorities of a significant cybersecurityThe activities needed to protect network and information systems, their users and other persons affected by cyber threats. NIS2 and the CRA adopt this Cybersecurity Act definition, so it frames most EU cybersecurity obligations. risk. If the manufacturer has ceased operations, inform the authorities without undue delay and, as far as possible, users.
Technical documentation
Draw up technical documentation with at least the Annex VII elements before placing the product on the market and update it as needed, at least during the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years.. Keep it and the EU declaration available to the market surveillance authorities for at least 10 years after the product is placed on the market, or for the support periodThe period during which a manufacturer must ensure that vulnerabilities of a product with digital elements are handled effectively and in line with the CRA vulnerability handling requirements. It must reflect the expected time of use and is generally at least five years. if longer.
Fines for misleading information
Applies to: Economic operators
Supplying incorrect, incomplete or misleading information to notified bodiesA conformity assessment body that a Member State has notified to the Commission as competent to carry out third-party assessment under EU legislation. Whether a manufacturer or provider must involve one depends on the applicable act, system or product classification and conformity assessment route; not every high-risk AI system requires a notified body. and market surveillance authorities in reply to a request carries fines of up to EUR 5 million or 1% of total worldwide annual turnover, whichever is higher.
Penalties and liability
Both cover it read the wording below
Fines for prohibited practices
Applies to: Operators of AI systems
Breach of the Article 5 prohibitions carries fines of up to EUR 35 million or 7% of total worldwide annual turnover, whichever is higher. For SMEs and start-ups the lower of the two applies.
Fines for operator obligations
Applies to: Operators of AI systems and notified bodies
Breach of the obligations of providers, authorised representatives, importers, distributors, deployers and notified bodies, or of Article 50 transparency, carries fines of up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher. For SMEs and small mid-caps the lower of the two applies.
Fines for misleading information
Applies to: Anyone supplying information to notified bodies or national competent authorities in reply to a request
Supplying incorrect, incomplete or misleading information to notified bodies or national authorities in reply to a request carries fines of up to EUR 7.5 million or 1% of total worldwide annual turnover, whichever is higher. For SMEs and small mid-caps the lower of the two applies.
Fines for general-purpose AI model providers Optional
Applies to: Providers of general-purpose AI models
The Commission may fine a provider that intentionally or negligently infringes the Regulation, fails to answer an information request correctly, or does not comply with a requested measure or evaluation access, up to 3% of annual total worldwide turnover or EUR 15 million, whichever is higher.
Fines for essential requirements and reporting
Applies to: Manufacturers
Breach of Annex I or of Articles 13 and 14 carries fines of up to EUR 15 million or 2.5% of total worldwide annual turnover, whichever is higher. Micro and small enterprises are not fined for missing the 24-hour early warning, and open-source software stewardsA legal person, other than a manufacturer, that systematically provides sustained support for the development of free and open-source software intended for commercial activities and ensures its viability, such as a foundation. The CRA gives stewards a lighter regime instead of full manufacturer obligations. are not fined for any infringement.
Fines for other operator obligations
Applies to: Economic operators and others subject to the listed articles
Breach of Articles 18 to 23, 28, 30(1) to (4), 31(1) to (4), 32(1) to (3), 33(5) and certain other listed articles carries fines of up to EUR 10 million or 2% of total worldwide annual turnover, whichever is higher.
Fines for misleading information
Applies to: Economic operators
Supplying incorrect, incomplete or misleading information to notified bodiesA conformity assessment body that a Member State has notified to the Commission as competent to carry out third-party assessment under EU legislation. Whether a manufacturer or provider must involve one depends on the applicable act, system or product classification and conformity assessment route; not every high-risk AI system requires a notified body. and market surveillance authorities in reply to a request carries fines of up to EUR 5 million or 1% of total worldwide annual turnover, whichever is higher.
Requirements are summarised in plain words; the official text prevails. Verdicts compare what the acts covered state, not how a supervisor applies them. Most topics are cross-referenced to NIST CSF 2.0 and ISO/IEC 27001:2022.
More Comparisons
Subject to more than one?
We run one control set for both, without doing the work twice. Ask what that saves you.