Map of the AI Act
Annex III: high-risk AI
The map shows what hangs off this object: who carries the duty, what it calls for, what it is evidenced by and what it rests on. Click an object and it becomes the new centre.
On a small screen the map reads as a path rather than as a network.
obligation
Annex III: high-risk AI
Classification route for standalone high-risk AI systems under Article 6(2) and Annex III.
based on
- Binding amended application calendarguidance
carries the duty
is affected
calls for
evidenced by
controlled by
supported by
in practice
explained in
- Article 6 has two separate routes to high-riskguidance · editorial reading
- Broadly positioned and general purpose AI systems: a disclaimer is not enoughguidance · editorial reading
- High-risk does not mean prohibited, and not high-risk does not mean permittedguidance · editorial reading
- Split and agentic architectures are assessed as a wholeguidance · editorial reading
- The Article 6(3) filter: four alternative conditions, to be read narrowlyguidance · editorial reading
- A human in the loop does not make a system low-riskguidance · editorial reading
- A human in the loop does not make a system low-riskguidance · editorial reading
- Improving is deliberately different from reviewingguidance · editorial reading
- Intended purpose is the anchor of classificationguidance · editorial reading
- First the definition question: is it an AI system at all?guidance · editorial reading
- Only the assessment of natural persons falls within these use casesguidance · editorial reading
- Preparatory or decisive: general input is allowed, a specific recommendation is notguidance · editorial reading
changed by
Read the map atin force now2 Feb 20252 Aug 202527 Jul 20262 Aug 20262 Dec 20262 Aug 20272 Dec 20272 Aug 2028
Read at 2 Aug 2027. What did not yet apply on that date is not shown here. GPAI models already on the market before 2 August 2025 must comply.