Written and maintained by CASRAI Editorial Board
Last updated
AI governance officers, medical device regulatory affairs teams, and research compliance directors searching for EU AI Act high risk rules must navigate the core compliance tier of Regulation (EU) 2024/1689 (the European Union Artificial Intelligence Act, published in the Official Journal on 12 July 2024 and entering into force on 1 August 2024). Unlike unacceptable-risk AI practices prohibited outright under Article 5 or general-purpose AI (GPAI) models regulated under Title VIII (see EU AI Act vs. California SB 53), EU AI Act high risk AI systems are permitted on the European Union market only if they satisfy the two-pathway classification rules of Article 6 and comply with the horizontal engineering, data governance, and conformity assessment obligations of Chapter III (Articles 8–49).
What Is High-Risk AI Under the EU AI Act? The Two Pathways of Article 6
Under EU AI Act Article 6, an artificial intelligence system is classified as high-risk through one of two distinct legal pathways, each carrying a different enforcement timeline and conformity assessment route:
| Classification Dimension | Pathway 1: Article 6(1) — Annex I Regulated Products | Pathway 2: Article 6(2) — Annex III Standalone Use Cases |
|---|---|---|
| Statutory Trigger | The AI system is a safety component of a product, or is itself a product, covered by Union harmonization legislation listed in Annex I, and is required to undergo a third-party conformity assessment under that Annex I law. | The AI system is deployed for one of the standalone high-risk use cases enumerated across the eight critical domains of Annex III (unless exempted by the narrow procedural derogation in Article 6(3)). |
| Healthcare & Research Examples | AI/ML Software as a Medical Device (SaMD) and diagnostic algorithms classified as Class IIa, IIb, or III under the EU Medical Devices Regulation (MDR, Regulation 2017/745) or Class B, C, or D under the In Vitro Diagnostic Regulation (IVDR, Regulation 2017/746). | AI systems used for emergency healthcare patient triage (Annex III, Point 5(d)), life and health insurance risk assessment and pricing (Point 5(c)), university admission/grading (Point 3), or recruitment/HR filtering (Point 4). |
| Conformity Assessment Route (Article 43) | Third-Party Notified Body Assessment: Integrated into the existing Notified Body conformity assessment required under the applicable Annex I legislation (e.g., MDR/IVDR Notified Body audit). | Internal Control (Self-Assessment per Annex VI) for Annex III Points 2–8, except Annex III Point 1 (Biometrics), which may require Notified Body involvement (Annex VII) if harmonized standards are not applied. |
| Enforcement Effective Date (Article 113) | 2 August 2027 (36 months after entry into force). | 2 August 2026 (24 months after entry into force). |
EU AI Act Annex III: The 8 Standalone High-Risk Domains and the Article 6(3) Derogation
For standalone software not regulated as a physical product or medical device under Annex I, EU AI Act Annex III high-risk use cases enumerate eight specific domains where algorithmic errors threaten fundamental rights, health, or safety:
- Biometrics (insofar as permitted by EU/national law): Remote biometric identification systems (excluding live public-space law-enforcement bans under Article 5), biometric categorization according to sensitive/protected attributes, and emotion recognition systems.
- Critical Infrastructure: AI systems intended to be used as safety components in the management and operation of critical digital infrastructure, road traffic, or the supply of water, gas, heating, or electricity.
- Education and Vocational Training: AI systems used to determine access or admission to educational institutions, evaluate learning outcomes, assess appropriate education levels, or monitor and detect prohibited student behavior during tests.
- Employment, Workers Management, and Access to Self-Employment: AI systems used for recruitment or selection (targeted job advertisements, CV screening, candidate evaluation) and for decisions affecting terms of work, promotion, termination, task allocation, or performance monitoring.
- Access to and Enjoyment of Essential Private and Public Services and Benefits: AI systems used by public authorities to evaluate eligibility for essential public assistance benefits and healthcare services; creditworthiness/credit-scoring systems; risk assessment and pricing in relation to natural persons in the case of life and health insurance; and AI systems evaluating and classifying emergency calls or dispatching/prioritizing emergency first-response services, including emergency healthcare patient triage systems.
- Law Enforcement: Polygraphs, crime-risk profiling, evidence reliability evaluation, and criminal analytics (where permitted by law).
- Migration, Asylum, and Border Control Management: Polygraphs, security/irregular migration risk assessments, and asylum/visa/residence application examination.
- Administration of Justice and Democratic Processes: AI systems intended to assist a judicial authority in researching and interpreting facts and the law, and systems intended to influence the outcome of an election or referendum.
The Article 6(3) Narrow Procedural Task Derogation
Under Article 6(3), an AI system listed in Annex III is not considered high-risk if it does not pose a significant risk of harm to the health, safety, or fundamental rights of natural persons by fulfilling one or more of four narrow conditions: (a) it is intended to perform a narrow procedural task, (b) it is intended to improve the result of a previously completed human activity, (c) it is intended to detect decision-making patterns or deviations from prior decision-making patterns without replacing or influencing the previously completed human assessment without proper human review, or (d) it is intended to perform a preparatory task to an assessment relevant for Annex III. However, Article 6(3) explicitly states that an AI system listed in Annex III shall always be considered high-risk if the AI system performs profiling of natural persons (within the meaning of GDPR Article 4(4)), and any provider claiming an Article 6(3) exemption must document its assessment prior to market placement and register the system in the EU database under Article 49(2).
Mandatory Compliance Checklist for High-Risk AI Providers (Articles 8–15)
Providers of high-risk AI systems (which researchers often cross-walk in EU AI Act vs. NIST AI RMF vs. ISO/IEC 42001 comparisons) must implement and document the following mandatory requirements under Chapter III, Section 2 before affixing CE marking:
- Article 9 — Continuous Risk Management System: Establish, implement, document, and maintain a continuous iterative risk management system across the entire lifecycle of the high-risk AI system, identifying known and reasonably foreseeable risks to health, safety, and fundamental rights under both intended use and reasonably foreseeable misuse.
- Article 10 — Data and Data Governance: Ensure training, validation, and testing datasets are subject to documented data governance practices covering design choices, data collection origin, annotation/labeling procedures, examination for possible biases likely to affect health/safety or lead to prohibited discrimination, and identification of relevant data gaps.
- Article 11 & Annex IV — Technical Documentation: Draw up comprehensive Annex IV technical documentation prior to market placement demonstrating compliance with Articles 8–15 for inspection by EU AI Act National Competent Authorities and Notified Bodies.
- Article 12 — Automatic Record-Keeping (Logging): Design the system with technical capabilities for the automatic recording of events (logs) over the lifetime of the system to trace input data, reference databases, operating periods, and the natural persons verifying results under Article 14(5).
- Article 13 — Transparency and Provision of Information to Deployers: Provide concise, complete, and clear Instructions for Use specifying the provider’s identity, system capabilities and known performance limitations, intended purpose, accuracy metrics across subgroups, and human oversight measures.
- Article 14 — Human Oversight: Design the system with human–machine interface tools enabling natural persons assigned by the deployer to understand the system’s capacities, guard against automation bias, interpret outputs, decide not to use or override the output, or immediately interrupt the system via a “stop” button.
- Article 15 — Accuracy, Robustness, and Cybersecurity: Achieve and declare appropriate levels of accuracy, technical redundancy/fail-safe plans, resilience against feedback loops ( in continuous learning systems), and protection against adversarial attacks (data poisoning, model poisoning, model evasion, and confidentiality attacks).
- Article 26 & Article 27 — Deployer Obligations and Fundamental Rights Impact Assessment (FRIA): Hospital systems, insurers, and public bodies deploying high-risk Annex III AI systems must follow the provider’s Instructions for Use, assign qualified human oversight personnel, retain automatic logs for at least 6 months, and—for public bodies or private entities providing public services (including healthcare and insurance under Annex III Points 5(b) and 5(c))—complete a Fundamental Rights Impact Assessment (FRIA) under Article 27 prior to first deployment.
Scientific Research and Open-Source Carve-Outs (Article 2)
Academic investigators should note the scope boundaries in Article 2(6) and 2(8): the EU AI Act does not apply to AI systems or models specifically developed and put into service for the sole purpose of scientific research and development, nor to research, testing, and development activities regarding AI systems prior to their being placed on the market or put into service (excluding testing in real-world conditions, which is governed by Article 60). However, once a university hospital or spin-out deploys an AI model into live clinical triage (Annex III Point 5(d)) or diagnostic patient care (Annex I MDR/IVDR), the high-risk obligations of Articles 8–15 apply in full, and the open-source exemption in Article 2(12) explicitly does not exempt high-risk AI systems.
