06 · High-Risk AI Classification
- Regulation (EU) 2024/1689 + Regulation (EU) 2026/1744 + Commission high-risk guidelines page
- Annex III high-risk Chapter III, Sections 1–3 obligations apply from 2 December 2027. Annex I high-risk Chapter III, Sections 1–3 obligations apply from 2 August 2028.
Annex III high-risk Chapter III, Sections 1–3 obligations apply from 2 December 2027. Annex I high-risk Chapter III, Sections 1–3 obligations apply from 2 August 2028.
Two tracks into high-risk classification (Article 6)
Track 1 — Annex I (regulated products)
AI systems that are a safety component of a product covered by existing EU product safety laws in Annex I (e.g. machinery, toys, medical devices, aviation, motor vehicles, rail, lifts) and that product requires third-party conformity assessment under those Annex I laws.
Regulation (EU) 2026/1744 narrows the definition of a safety component. The AI component must have an intended safety function that prevents or mitigates risks to the health and safety of people or property, or its failure or malfunction must endanger them. Integration into a regulated product does not by itself make an AI system a safety component.
→ Article 6(1) / Annex I high-risk Chapter III, Sections 1–3 obligations apply from 2 August 2028.
Track 2 — Annex III (use cases)
AI systems listed in specific use-case areas in Annex III. Currently 8 areas:
| Area | Examples |
|---|---|
| Biometrics (non-banned) | Remote biometric ID, biometric categorisation, emotion recognition (not banned) |
| Critical infrastructure | AI managing road traffic, water/gas/electricity/heating, digital infrastructure |
| Education & vocational training | Admission decisions, exam scoring, student behaviour monitoring |
| Employment & HR | CV screening, interview analysis, promotion/termination, performance monitoring |
| Essential public & private services | Benefits eligibility, credit scoring, emergency call dispatch, insurance risk pricing |
| Law enforcement | Victim risk assessment, evidence reliability evaluation, offending risk (non-solo profiling) |
| Migration & border control | Visa/asylum examination, border crossing irregular migration risk |
| Justice & democratic processes | Court ruling assistance, election/voting influence tools |
→ Article 6(2) / Annex III high-risk Chapter III, Sections 1–3 obligations apply from 2 December 2027.
Exceptions — Annex III systems that are not high-risk
An AI system listed in Annex III is not classified high-risk if it:
- Performs a narrow procedural task
- Improves the result of a previously completed human activity
- Detects decision-making patterns without replacing human judgment
- Performs a preparatory task to an assessment (i.e. not the assessment itself)
Exception does not apply if the system profiles individuals — automated processing of personal data to assess work performance, economic situation, health, preferences, reliability, behaviour, location, or movement. Profiling always makes the system high-risk.
Providers whose system falls under Annex III but believe it's not high-risk must document that assessment before market placement. They must also register the system in the EU database under Article 49(2), using the reduced Annex VIII fields introduced by Regulation (EU) 2026/1744.
Obligations on high-risk AI providers
Once classified high-risk, providers must:
| Obligation | Article |
|---|---|
| Risk management system (documented, lifecycle-long) | Art. 9 |
| Data governance (relevant, representative, error-free datasets) | Art. 10 |
| Technical documentation (Annex IV format) | Art. 11 |
| Automatic event logging / record-keeping | Art. 12 |
| Instructions for use (for deployers) | Art. 13 |
| Human oversight design | Art. 14 |
| Accuracy, robustness, cybersecurity | Art. 15 |
| Quality management system | Art. 17 |
Obligations on high-risk AI deployers
- Implement human oversight as instructed by providers
- Monitor system performance in production
- Report serious incidents and malfunctions to providers and national authorities
- Conduct fundamental rights impact assessments (for certain deployers — public bodies, banks, insurers)
Conformity assessment
See 07 · Conformity Assessment for the full process.
- For Annex I (product safety) systems: third-party notified body assessment
- For most Annex III systems: self-assessment (internal conformity assessment) is sufficient
- Exception: biometric identification systems (public-space RBI, remote biometric ID) require third-party assessment
After assessment: affix CE marking and issue EU declaration of conformity.
Post-market: register in the EU database (for Annex III systems used by or against individuals). Public authorities deploying high-risk AI must register before deployment.
Draft classification guidelines (status: 7 September 2026)
The Commission published draft guidelines clarifying the classification rules, including the Article 6 exceptions, and opened a further targeted consultation that closed on 23 July 2026. The Commission page says feedback will be incorporated before final adoption. No final guidelines were located in the 7 September 2026 audit. URL: https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-high-risk-systems
Key articles and annexes
| Topic | Reference |
|---|---|
| Classification rules | Art. 6 |
| Annex I (regulated product laws) | Annex I |
| Annex III (use cases) | Annex III |
| Risk management | Art. 9 |
| Data governance | Art. 10 |
| Technical documentation | Art. 11 + Annex IV |
| Record-keeping | Art. 12 |
| Transparency to deployers | Art. 13 |
| Human oversight | Art. 14 |
| Accuracy/robustness/cybersecurity | Art. 15 |
| Quality management | Art. 17 |
| Conformity assessment | Arts. 43–49 |
| EU database | Art. 71 |
| Fundamental rights impact assessment | Art. 27 |