EU Artificial Intelligence Act

Screen every AI capability offered in the EU, or whose output is used in the EU, against the Article 5 prohibition list (manipulation and deception, exploitation of vulnerabilities, social scoring, individual criminal-offence prediction from profiling, untargeted facial-image scraping, emotion recognition at work and in education, biometric categorisation inferring protected characteristics, real-time remote biometric identification for law enforcement, and from 2 Dec 2026 the generation of non-consensual intimate imagery and child sexual abuse material). Remove or technically block any hit before placing the system on the market, and keep the assessment on file.

Applies when

Any provider or deployer placing an AI system on the EU market, putting it into service in the EU, or whose system output is used in the EU, regardless of establishment.

Deadline: Before placing on the EU market; the two new prohibitions in Art. 5(1)(ba) and (bb) apply from 2 Dec 2026.

Applies from: 2025-02-02

Evidence to retain

• · Article 5 prohibition screening matrix mapped to product features, with conclusions
• · Technical removal or blocking design for any hit, plus pre-release verification records
• · Signed review minutes from legal and product owners on prohibited practices
• · Description of misuse safeguards for synthetic-content products (safety filters, prompt blocking, abuse-report handling)

Estimated first-time effort: 5–15 person-days (estimate, not a commitment)

Providers and deployers must take measures to build the AI literacy of their own staff and of persons operating AI systems on their behalf, taking account of technical knowledge, experience, education and context of use. As amended by the AI Omnibus, the duty does not require guaranteeing any specific literacy level for any individual, but evidence of training is still expected.

Applies when

All providers and deployers of AI systems in the EU, including Chinese companies operating through an EU subsidiary or distributor.

Applies from: 2025-02-02

Evidence to retain

• · AI literacy training syllabus and materials
• · Role-based attendance list and completion records
• · Assessment or knowledge-check records evidencing effectiveness

Estimated first-time effort: 3–8 person-days (estimate, not a commitment)

Draw up and keep up to date model technical documentation covering the training and testing process and evaluation results, containing at least the Annex XI elements, and provide it to the AI Office and national competent authorities on request. Separately, supply downstream AI system providers with information and documentation containing at least the Annex XII elements so they can understand the model’s capabilities and limitations and meet their own obligations. Models released under a genuinely free and open-source licence with public weights, architecture and usage information are exempt from both, unless the model has systemic risk.

Applies when

Providers placing a general-purpose AI model on the EU market, including Chinese companies serving EU customers via API, cloud marketplaces, model-hosting platforms or bundled with a product.

Deadline: Applies from 2 Aug 2025; models already placed on the EU market before 2 Aug 2025 must be brought into compliance by 2 Aug 2027 (Art. 111(3)).

Applies from: 2025-08-02

Evidence to retain

• · Annex XI-compliant model technical documentation (training and testing process, evaluation results, compute accounting)
• · Model card and integration documentation for downstream providers covering the Annex XII elements
• · Document version-control log mapped one-to-one to model releases
• · Licence text and public links to weights, architecture and usage information where the open-source exemption is claimed

Estimated first-time effort: 15–40 person-days (estimate, not a commitment)

Draw up and publish a sufficiently detailed summary of the content used for training, using the official AI Office template. It must set out the composition of data sources (including large datasets and main domain names) and how data was obtained and processed, so that rightsholders and other parties with legitimate interests can exercise their rights under Union law. The summary must match the model as actually trained and be updated with each release.

Applies when

All providers placing a GPAI model on the EU market. The Art. 53(2) open-source exemption covers only points (a) and (b) and does not exempt this obligation.

Applies from: 2025-08-02

Evidence to retain

• · Public training-content summary completed on the official template, with the published URL archived
• · Register of data sources and the licence or legal basis for each (proprietary, licensed, web-crawled, third-party datasets)
• · Internal reconciliation record showing the summary matches the data actually used
• · Change history of the summary across model releases

Estimated first-time effort: 10–25 person-days (estimate, not a commitment)

Put in place and operate a policy to comply with Union copyright law, in particular identifying and respecting rights reservations expressed under Article 4(3) of Directive (EU) 2019/790 using state-of-the-art technologies (for example robots.txt and machine-readable opt-out signals). The policy must cover crawling, dataset procurement, fine-tuning data and synthetic-data pipelines.

Applies when

All providers placing a GPAI model on the EU market, including Chinese companies that train outside the EU but serve the model or its API into the EU. Open-source models are not exempt.

Applies from: 2025-08-02

Evidence to retain

• · Copyright compliance policy text with approval record
• · Crawler and data-collection configuration (robots.txt compliance, opt-out signal detection rules) and execution logs
• · Dataset procurement contracts and chain-of-rights evidence
• · Log of rightsholder complaints and takedown requests and how they were handled

Estimated first-time effort: 10–25 person-days (estimate, not a commitment)

Continuously account for cumulative training compute. Under Art. 51(2), a GPAI model trained with cumulative compute greater than 10^25 floating-point operations is presumed to have high-impact capabilities and is therefore classified as a model with systemic risk. Notify the Commission within two weeks of meeting, or knowing it will meet, that condition, with supporting evidence. A provider may submit substantiated arguments with the notification that, exceptionally, the model does not present systemic risk.

Applies when

Chinese companies training large foundation models and serving them into the EU. The compute accounting covers pre-training, synthetic-data generation and fine-tuning, i.e. all activities intended to enhance capabilities. Because Art. 51(3) empowers the Commission to amend the threshold, re-check the current delegated act.

Deadline: Within two weeks of meeting, or becoming aware that it will meet, the condition.

Applies from: 2025-08-02

Evidence to retain

• · Compute accounting methodology and a per-run FLOP ledger
• · Threshold-crossing determination record with date and accountable owner
• · Notification submitted to the Commission and the acknowledgement received
• · Substantiated arguments and supporting evidence where exemption from systemic-risk classification is claimed

Estimated first-time effort: 5–12 person-days (estimate, not a commitment)

In addition to Articles 53 and 54, providers of models with systemic risk must perform model evaluation under standardised state-of-the-art protocols, including documented adversarial testing; assess and mitigate possible Union-level systemic risks; keep track of, document and report serious incidents and corrective measures to the AI Office without undue delay (and to national competent authorities where appropriate); and ensure an adequate level of cybersecurity for the model and its physical infrastructure.

Applies when

Providers of GPAI models classified as having systemic risk, including models designated ex officio by the Commission.

Applies from: 2025-08-02

Evidence to retain

• · Model evaluation protocol and evaluation reports, including adversarial testing and red-teaming records
• · Systemic-risk identification and mitigation register with residual-risk conclusions
• · Serious-incident log and records of reports filed with the AI Office
• · Security controls for model weights and training infrastructure, with audit evidence

Estimated first-time effort: 30–80 person-days (estimate, not a commitment)

Before placing a GPAI model on the Union market, a provider established in a third country must appoint, by written mandate, an authorised representative established in the Union. The representative verifies that the Annex XI technical documentation has been drawn up and that the Article 53 (and where applicable Article 55) obligations have been met, keeps a copy of that documentation and the provider’s contact details at the disposal of the AI Office for 10 years after the model is placed on the market, supplies information on reasoned request, and cooperates with regulators on that model. Free and open-source models with public weights, architecture and usage information are exempt unless they present systemic risk.

Applies when

Companies established in China placing a GPAI model on the EU market. This is the pre-market condition most often overlooked by Chinese foundation-model vendors entering the EU.

Deadline: The mandate must be in place before the model is placed on the Union market.

Applies from: 2025-08-02

Evidence to retain

• · Written mandate covering all tasks listed in Art. 54(3)
• · Proof of the representative’s establishment in the EU and its contact details
• · Arrangement for holding the technical documentation at the representative, with the 10-year retention period stated
• · Records of routine liaison with the representative and handling of regulator correspondence

Estimated first-time effort: 5–15 person-days (estimate, not a commitment)

AI systems intended to interact directly with natural persons (chatbots, voice assistants, customer-service and sales agents) must be designed so that users are informed they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person. The information must be given clearly and distinguishably at the latest at the first interaction and must meet accessibility requirements.

Applies when

Providers of conversational products used by EU users, including Chinese companies embedding via API or SDK into EU customers’ products and those whose output is used in the EU.

Applies from: 2026-08-02

Evidence to retain

• · In-product AI disclosure copy and interface screenshots for each language
• · Technical description of the first-interaction trigger logic
• · Accessibility compliance check record

Estimated first-time effort: 3–10 person-days (estimate, not a commitment)

Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, with technical solutions that are effective, interoperable, robust and reliable. Systems performing an assistive function for standard editing, or that do not substantially alter the deployer’s input data or its semantics, are out of scope. Compliance may be demonstrated by adhering to the Code of Practice on Transparency of AI-generated Content, the voluntary code published on 10 June 2026; non-adherents must demonstrate compliance by alternative equivalently adequate means.

Applies when

Chinese companies offering text-to-image, text-to-video, voice synthesis or text generation into the EU. Note that systems already placed on the EU market before 2 Aug 2026 must comply with Art. 50(2) by 2 Dec 2026 under the new Art. 111(4) inserted by the AI Omnibus.

Deadline: Applies to new systems from 2 Aug 2026; legacy systems placed on the market before 2 Aug 2026 must comply by 2 Dec 2026.

Applies from: 2026-08-02

Evidence to retain

• · Technical description of the marking scheme (watermarking, metadata, provenance credentials) and its coverage
• · Test report on effectiveness, robustness and interoperability of the marking
• · Monitoring records of marking coverage and missed-marking incidents
• · Proof of signature of the Code of Practice on Transparency of AI-generated Content, or the equivalently-adequate-means justification if not signed

Estimated first-time effort: 15–40 person-days (estimate, not a commitment)

Deployers of emotion recognition or biometric categorisation systems must inform the natural persons exposed to them and process personal data in line with the GDPR. Deployers generating or manipulating image, audio or video content constituting a deep fake must disclose that it is artificially generated or manipulated; the same applies to AI-generated or manipulated text published to inform the public on matters of public interest, unless the content underwent human review with a natural or legal person holding editorial responsibility. For evidently artistic, creative, satirical or fictional works the disclosure may be limited to an appropriate manner that does not hamper enjoyment of the work.

Applies when

Deployers using such systems in the EU, and third-country deployers whose system output is used in the EU, including the EU marketing, customer-service and content teams of Chinese companies.

Applies from: 2026-08-02

Evidence to retain

• · Deepfake disclosure labelling standard and sampled evidence of published content
• · Notice copy and delivery records for emotion recognition or biometric categorisation contexts
• · GDPR lawful basis and cross-reference to the processing records for those activities

Estimated first-time effort: 5–15 person-days (estimate, not a commitment)

Assess each system against Art. 6(1) (a safety component of, or itself, a product covered by the Union harmonisation legislation in Annex I) and Art. 6(2) (the Annex III use cases). The AI Omnibus inserted Art. 6(1a)-(1c): systems used solely for non-safety-related user assistance, performance optimisation, service efficiency, automation, convenience or quality control do not qualify as safety components, but systems whose failure or malfunction would endanger health and safety do. Where a provider concludes under Art. 6(3) that an Annex III system is not high-risk, it must register itself and that system in the EU database before placing it on the market.

Applies when

All providers placing AI systems on the EU market, especially embodied-AI and robotics, autonomous driving, healthcare and industrial inspection vendors whose products often fall under both the Annex I product legislation and the Annex III use-case list.

Deadline: The two application dates differ and must be scheduled separately by classification route: systems high-risk under Art. 6(2) and Annex III apply from 2 Dec 2027; systems high-risk under Art. 6(1) and Annex I apply from 2 Aug 2028 (Art. 113, third paragraph, point (c), as amended by the AI Omnibus). The other high-risk obligations in this corpus carry the earlier date, 2 Dec 2027, as their effectiveFrom; companies on the Annex I route should shift them to 2 Aug 2028.

Applies from: 2027-12-02

Evidence to retain

• · Per-system high-risk classification matrix comparing against Art. 6(1) and 6(2) with conclusions
• · Safety-component determination applying Art. 6(1a)-(1c)
• · Documented reasoning where a system is concluded not to be high-risk under Art. 6(3)
• · EU database registration receipt where registration under Art. 49(2) applies

Estimated first-time effort: 5–15 person-days (estimate, not a commitment)

Establish, implement, document and maintain a continuous iterative risk management process across the system lifecycle: identify and analyse known and reasonably foreseeable risks to health, safety and fundamental rights under intended use; estimate risks that may arise under intended use and reasonably foreseeable misuse; evaluate further risks using post-market monitoring data under Art. 72; and adopt targeted risk management measures so that each residual risk and the overall residual risk are judged acceptable.

Applies when

Providers of high-risk AI systems, including products developed in China and placed on the EU market through an EU importer or distributor.

Applies from: 2027-12-02

Evidence to retain

• · Risk management plan and risk register including reasonably foreseeable misuse scenarios
• · Risk assessment and mitigation records with residual-risk acceptability conclusions
• · Records of periodic systematic review and updating
• · Evidence that post-market monitoring data feeds back into the risk assessment

Estimated first-time effort: 20–50 person-days (estimate, not a commitment)

For high-risk systems using model-training techniques, training, validation and testing datasets must meet the Art. 10(2)-(4) quality criteria: design choices, data origin and collection, preparation, assumptions, assessment of sufficiency, examination for and mitigation of bias, handling of data gaps, and representativeness for the intended persons and the geographic and contextual setting. For systems not using training techniques, these apply only to testing datasets. The AI Omnibus deleted the former Art. 10(5) and inserted Art. 4a, permitting processing of special categories of personal data for bias detection and correction under strict conditions.

Applies when

Providers of high-risk AI systems. For Chinese vendors the most common gap is inability to evidence data provenance and geographic representativeness; the data distribution of the EU deployment setting must be explicitly assessed.

Applies from: 2027-12-02

Evidence to retain

• · Dataset card or datasheet covering provenance, collection process, preparation and labelling rules
• · Representativeness and data-gap assessment report addressing the EU target population
• · Bias examination and mitigation records
• · Where Art. 4a is relied on, the necessity justification, safeguards applied and deletion records

Estimated first-time effort: 20–60 person-days (estimate, not a commitment)

Draw up the technical documentation before the system is placed on the market or put into service and keep it up to date. It must contain at least the Annex IV elements in a clear and comprehensive form allowing national competent authorities and notified bodies to assess compliance. As amended by the AI Omnibus, Art. 11(1) allows SMEs including start-ups and small mid-cap companies (SMCs) to use a simplified form published by the Commission, which notified bodies must accept.

Applies when

Providers of high-risk AI systems. The documentation must be updated at every substantial change and be retrievable on demand during conformity assessment and market surveillance.

Applies from: 2027-12-02

Evidence to retain

• · Complete Annex IV technical documentation, or the simplified SME/SMC form
• · Document version log and change-tracking table
• · Post-market monitoring plan, which forms part of the technical documentation
• · Designation of the person accountable for retaining and producing the documentation

Estimated first-time effort: 20–50 person-days (estimate, not a commitment)

High-risk systems must technically allow automatic recording of events over their lifetime, with logging capabilities sufficient to identify situations that may cause the system to present a risk within the meaning of Art. 79(1) or to undergo a substantial modification, to support post-market monitoring under Art. 72, and to enable deployer monitoring under Art. 26(5). Providers must keep the automatically generated logs under their control for at least six months unless other Union or national law provides otherwise; deployers must do the same for logs under their control. For remote biometric identification systems the logs must at minimum record the start and end of each use, the reference database, the input data that produced a match, and the identity of the persons verifying results.

Applies when

Providers and deployers of high-risk AI systems. Any tension between the retention period and the GDPR storage-limitation principle must be addressed explicitly in the retention policy.

Applies from: 2027-12-02

Evidence to retain

• · Log schema design and retention policy documenting the six-month minimum
• · Description of log integrity and tamper-resistance controls
• · Sample log export used in a regulator-request drill
• · Records of data minimisation and access control for personal data in logs

Estimated first-time effort: 10–25 person-days (estimate, not a commitment)

High-risk systems must be designed so that their operation is sufficiently transparent for deployers to interpret and properly use the output, and must be accompanied by instructions for use in an appropriate digital or other format giving deployers the information they need to meet their own obligations, including capabilities and limitations, declared accuracy metrics, intended purpose, known reasonably foreseeable misuse, and how human oversight measures are to be implemented.

Applies when

Providers of high-risk AI systems. Instructions must be provided in a language easily understood in the deployer’s Member State.

Applies from: 2027-12-02

Evidence to retain

• · Released instructions for use in each target language
• · Basis of calculation for the declared accuracy metrics and limitations
• · Deployer training materials and delivery records

Estimated first-time effort: 8–20 person-days (estimate, not a commitment)

Systems must be designed, including through appropriate human-machine interface tools, so that natural persons can effectively oversee them while in use, in order to prevent or minimise risks to health, safety and fundamental rights under intended use and reasonably foreseeable misuse. Oversight measures must be commensurate with the risks, level of autonomy and context of use, and may be built into the system by the provider or identified by the provider for the deployer to implement. They must enable the overseeing persons to understand capacities and limitations, remain aware of automation bias, correctly interpret output, and decide not to use the output or to interrupt or stop the system.

Applies when

Providers and deployers of high-risk AI systems. Deployers must assign human oversight to natural persons who have the necessary competence, training, authority and support.

Applies from: 2027-12-02

Evidence to retain

• · Human oversight design document covering interfaces, intervention points and stop mechanisms
• · Job descriptions, authorisations and training records for overseeing staff
• · Sample operational logs of human intervention and override of outputs
• · Product description of measures addressing automation bias

Estimated first-time effort: 10–30 person-days (estimate, not a commitment)

Systems must achieve an appropriate level of accuracy, robustness and cybersecurity and perform consistently in those respects throughout their lifecycle. Accuracy levels and relevant metrics must be declared in the instructions for use. Systems must be as resilient as possible to errors, faults or inconsistencies within the system or its environment, which may be achieved through technical redundancy, backup or fail-safe plans. Systems that continue to learn after being placed on the market must be developed so as to eliminate or reduce as far as possible the risk of biased outputs influencing future inputs (feedback loops), and must be resilient to unauthorised third-party attempts to alter their use, outputs or performance. The AI Omnibus inserted Art. 42(3): high-risk systems within the scope of the Cyber Resilience Act that satisfy its Art. 12(1) conditions are deemed to comply with Art. 15.

Applies when

Providers of high-risk AI systems. Embodied-AI and autonomous-driving vendors should pay particular attention to feedback-loop risk from online learning and to fail-safe design.

Applies from: 2027-12-02

Evidence to retain

• · Accuracy and robustness test plan and reports with metric definitions
• · Design description of redundancy and fail-safe arrangements
• · Controls and test records against adversarial attacks (data poisoning, model poisoning, adversarial examples, model evasion)
• · Feedback-loop monitoring and mitigation records for continuously learning systems

Estimated first-time effort: 20–50 person-days (estimate, not a commitment)

Put in place a quality management system documented systematically in written policies, procedures and instructions, covering at least: a regulatory compliance strategy including conformity assessment and change management; design and design-verification techniques; development and quality-assurance techniques; examination, test and validation procedures before, during and after development and their frequency; technical specifications and standards applied, with alternative means where harmonised standards are not fully applied; and data management systems and procedures. As amended, Art. 17(2) requires implementation proportionate to the provider’s organisational size, in particular for SMEs and SMCs, without lowering the required rigour or level of protection.

Applies when

Providers of high-risk AI systems. Companies already certified to ISO 9001 or ISO/IEC 42001 can map onto those systems but must close gaps against Art. 17 item by item.

Applies from: 2027-12-02

Evidence to retain

• · Quality management system manual and index of procedures
• · Mapping table from each Art. 17(1) item to existing system documents
• · Change management and release approval records
• · Internal audit and management review records

Estimated first-time effort: 25–60 person-days (estimate, not a commitment)

Determine the conformity assessment route under Art. 43. For Annex III point 1 (biometrics) systems where harmonised standards or common specifications have been applied, the provider may choose Annex VI internal control or Annex VII with notified body involvement; where those standards are absent or only partly applied, Annex VII is mandatory. Annex III points 2 to 8 follow Annex VI internal control. Systems covered by the Annex I Section A harmonisation legislation follow that legislation’s procedure, with the Chapter III Section 2 requirements integrated into it. On completion, draw up a written, machine-readable, physically or electronically signed EU declaration of conformity per system with the Annex V content, keep it at the disposal of national competent authorities for 10 years after placing on the market or putting into service, and affix the CE marking (a digital CE marking for digitally provided systems).

Applies when

Providers of high-risk AI systems. Chinese vendors should plan notified-body scheduling and English-language documentation well in advance; the Annex VII route typically takes multiple quarters.

Applies from: 2027-12-02

Evidence to retain

• · Rationale for the chosen conformity assessment route, including the status of harmonised standards and common specifications
• · Notified body certificate or internal-control assessment report
• · EU declaration of conformity with the Annex V content and the 10-year retention arrangement
• · Evidence of where the CE marking is affixed and of accessibility of the digital CE marking

Estimated first-time effort: 30–90 person-days (estimate, not a commitment)

Before placing on the market or putting into service a high-risk AI system listed in Annex III (except those under point 2), the provider or its authorised representative must register itself and the system in the EU database referred to in Art. 71; systems concluded not to be high-risk under Art. 6(3) must also be registered. Annex III point 2 systems are registered at national level. Systems under points 1, 6 and 7 in the areas of law enforcement, migration, asylum and border control are registered in a secure non-public section of the database.

Applies when

Providers of Annex III high-risk systems and their EU authorised representatives. The registry is publicly searchable and is one of the first items EU customers check in due diligence.

Deadline: Registration must be completed before placing on the market or putting into service. The AI Omnibus deleted points 7 and 9 of Annex VIII Section B, reducing the information to be registered for systems concluded not to be high-risk under Art. 6(3).

Applies from: 2027-12-02

Evidence to retain

• · EU database registration receipt and a screenshot of the entry
• · Consistency check between the registered information and the technical documentation
• · Update records for any change to the registered information

Estimated first-time effort: 3–10 person-days (estimate, not a commitment)

Before making a high-risk AI system available on the Union market, a provider established in a third country must appoint, by written mandate, an authorised representative established in the Union. The representative verifies that the EU declaration of conformity and technical documentation have been drawn up and that an appropriate conformity assessment has been carried out; keeps the provider’s contact details, a copy of the declaration, the technical documentation and any notified body certificate at the disposal of competent authorities for 10 years after the system is placed on the market or put into service; supplies on reasoned request all information and documentation needed to demonstrate conformity, including access to logs under the provider’s control; and cooperates with competent authorities.

Applies when

Companies established in China placing high-risk AI systems on the EU market. This is a separate obligation from the GPAI authorised representative; one does not substitute for the other.

Deadline: The mandate must be in place before the high-risk system is made available on the Union market.

Applies from: 2027-12-02

Evidence to retain

• · Written mandate covering all tasks in Art. 22(3)
• · Index of compliance documents held by the representative and the retention period
• · Contingency plan for the representative terminating the mandate

Estimated first-time effort: 5–15 person-days (estimate, not a commitment)

Establish and document a post-market monitoring system proportionate to the nature of the AI technologies and the system’s risks, actively and systematically collecting, documenting and analysing performance data from deployers or other sources to evaluate continuous compliance with the Chapter III Section 2 requirements and, where relevant, interaction with other AI systems. The system must be based on a post-market monitoring plan that forms part of the Annex IV technical documentation. The AI Omnibus removed the mandatory implementing-act template and instead requires the Commission to publish guidance, including a voluntary template, by 2 Sep 2027.

Applies when

Providers of high-risk AI systems. Chinese vendors should write performance-data feedback duties into contracts with EU distributors and integrators, without which the monitoring system cannot function.

Applies from: 2027-12-02

Evidence to retain

• · Post-market monitoring plan as part of the technical documentation
• · Description of data collection channels and metric definitions
• · Periodic monitoring analysis reports and corrective-action tracking
• · Contractual clauses with deployers and distributors on performance-data feedback

Estimated first-time effort: 15–35 person-days (estimate, not a commitment)

Providers of high-risk AI systems must report serious incidents to the market surveillance authority of the Member State where the incident occurred: immediately after establishing a causal link, or the reasonable likelihood of one, between the system and the incident, and in any event no later than 15 days after the provider or deployer becomes aware; no later than 2 days for a widespread infringement or a serious incident as defined in Art. 3(49)(b); and no later than 10 days in the event of a person’s death. An incomplete initial report may be filed first and completed later. After reporting, the provider must investigate without delay, including a risk assessment of the incident and corrective action, and must not alter the system in a way that could affect the subsequent assessment of causes before informing the competent authorities.

Applies when

Providers of high-risk AI systems, with the rules applying mutatis mutandis to deployers unable to reach the provider. The incident response chain must connect the China headquarters with the EU authorised representative and distributors, otherwise the 2-day deadline cannot be met.

Deadline: 15 days generally; 2 days for widespread infringement or Art. 3(49)(b) incidents; 10 days where a person has died.

Applies from: 2027-12-02

Evidence to retain

• · Serious-incident criteria and tiered response playbook mapped to the 2, 10 and 15-day deadlines
• · Incident register and submission acknowledgements
• · Incident investigation report with root-cause analysis and corrective actions
• · Records of evidence preservation and of the notify-before-modifying process

Estimated first-time effort: 8–20 person-days (estimate, not a commitment)

Deployers must take appropriate technical and organisational measures to use the system in line with the instructions for use; assign human oversight to natural persons with the necessary competence, training and authority; ensure input data is relevant and sufficiently representative to the extent they control it; monitor operation per the instructions and, where they have reason to believe the system presents a risk within the meaning of Art. 79(1), without undue delay inform the provider or distributor and the market surveillance authority and suspend use; on identifying a serious incident, inform first the provider, then the importer or distributor and the market surveillance authorities; keep logs under their control for at least six months; inform workers’ representatives and affected workers before putting a high-risk system into service at the workplace; and, where applicable, use the Art. 13 information to carry out a GDPR Art. 35 data protection impact assessment.

Applies when

Deployers using high-risk AI systems in the EU, including EU subsidiaries of Chinese groups using third-party AI tools for recruitment, performance management or credit assessment.

Applies from: 2027-12-02

Evidence to retain

• · Operating procedures derived from the instructions for use, plus approvals for any deviation
• · Authorisation and training records for human-oversight roles
• · Input data quality control records
• · Notice to workers’ representatives and proof of delivery where deployed at the workplace

Estimated first-time effort: 10–25 person-days (estimate, not a commitment)

The FRIA has a narrow, specific scope: before deploying an Art. 6(2) high-risk system (other than Annex III point 2 critical infrastructure), deployers that are bodies governed by public law or private entities providing public services, and deployers of Annex III points 5(b) and 5(c) systems (creditworthiness assessment and risk assessment and pricing for life and health insurance), must assess the impact on fundamental rights. The assessment covers the deployer’s processes, the period and frequency of use, the categories of persons and groups likely to be affected, the specific risks of harm, the implementation of human oversight measures, and the measures to be taken if risks materialise including internal governance and complaint mechanisms. Results must be notified to the market surveillance authority using the AI Office questionnaire template. It applies to first use and must be updated when elements change. The AI Omnibus allows cross-referencing to, or incorporation of, the GDPR Art. 35 DPIA.

Applies when

Note the boundary: most commercial deployers do not owe a FRIA. The scenarios most likely to catch Chinese groups are an EU subsidiary performing consumer credit scoring or life and health insurance pricing, or deploying an Annex III system as a provider of public services.

Deadline: Before the first use of the high-risk system, and updated whenever the elements change.

Applies from: 2027-12-02

Evidence to retain

• · FRIA applicability determination explaining why it is or is not triggered
• · Full FRIA report covering the six elements in Art. 27(1)
• · Notification to the market surveillance authority and the submitted questionnaire acknowledgement
• · Cross-reference table linking the FRIA to the DPIA

Estimated first-time effort: 8–20 person-days (estimate, not a commitment)

Where a third party substantially modifies a system or changes its intended purpose and becomes a new provider, the initial provider ceases to be the provider of that specific system but must cooperate closely with the new provider, making available the necessary information and reasonably expected technical access and assistance, including technical documentation sufficient to assess compliance with Art. 16, information on known limitations and failure modes, and targeted technical access for testing and validation. This does not apply where the initial provider has clearly specified that its system is not to be changed into a high-risk AI system. In addition, third parties supplying AI systems, models, tools, services, components or processes used or integrated in a high-risk system must specify by written agreement the information, capabilities, technical access and assistance the high-risk provider needs; this does not apply to tools, services, processes or components other than GPAI models made publicly available under a free and open-source licence. The AI Omnibus brought breaches of Art. 25(2) and (4) within the Art. 99(4) fine bracket.

Applies when

Chinese companies supplying foundation models, vision or speech components, robotics control stacks or MLOps tooling to EU customers: even when not themselves a high-risk provider, downstream integration creates cooperation and written-agreement duties.

Applies from: 2027-12-02

Evidence to retain

• · Template written agreement with downstream integrators covering the Art. 25(4) elements
• · Technical documentation and statements of known limitations and failure modes delivered downstream
• · Records of technical access granted for testing and validation
• · Where the carve-out is relied on, the explicit product-documentation statement that the system is not to be changed into a high-risk AI system

Estimated first-time effort: 8–20 person-days (estimate, not a commitment)