Annex III is the list that decides whether the EU AI Act's heaviest obligations apply to you. Eight categories, each naming specific uses. If your system does one of the things on this list, it is presumptively high-risk - and then Article 6(3) is the only way back out.
Being in an Annex III area is not the end of the analysis. Article 6(3) can remove a system from high-risk even inside a listed area - unless it profiles natural persons, in which case the exit is closed absolutely. How the exemption works, or run one system through the free high-risk checker.
The eight categories
| # | Area | What it covers | Who this usually catches |
|---|---|---|---|
| 1 | Biometrics | Remote biometric identification; biometric categorisation by sensitive or protected attributes; emotion recognition | Identity verification, access control, retail analytics |
| 2 | Critical infrastructure | Safety components in digital infrastructure, road traffic, and water, gas, heating or electricity supply | Industrial control, grid and traffic systems |
| 3 | Education & vocational training | Admission and access decisions; evaluating learning outcomes; assessing educational level; monitoring for prohibited behaviour during tests | EdTech, assessment platforms, proctoring |
| 4 | Employment & worker management | Recruitment, advertising vacancies, screening and evaluating candidates; promotion, termination, task allocation, performance monitoring | HR tech, ATS, workforce analytics - detail here |
| 5 | Essential private & public services | Public benefits and healthcare eligibility; creditworthiness and credit scoring (fraud detection excluded); life and health insurance risk pricing; emergency call triage and dispatch | Fintech, insurtech, govtech - detail here |
| 6 | Law enforcement | Victim risk assessment; polygraph-equivalent tools; evidence reliability; offending and re-offending risk; profiling of natural persons | Police tech, forensic tooling |
| 7 | Migration, asylum & border control | Polygraph-equivalent tools; risk assessment of persons entering; assistance examining asylum and visa applications; detection and identification of persons | Border systems, immigration case tooling |
| 8 | Justice & democratic processes | Assisting judicial authorities in researching and interpreting facts and law; systems intended to influence election outcomes or voting behaviour | Legal tech aimed at courts, civic and political tooling |
Categories 1, 6 and 7 apply only in so far as the use is permitted under applicable Union or national law. Several of the uses adjacent to them are not high-risk at all but prohibited outright under Article 5 - untargeted facial scraping, social scoring, emotion inference in workplaces and schools. Prohibited is a different and worse category than high-risk.
The two questions that decide most cases
Is a natural person the subject of the evaluation? This runs through the whole Annex. Scoring a company's credit is not scoring a person. Parsing a CV is not evaluating the candidate. Analysing transactions is not scoring the customer. The moment the human becomes the thing being assessed, you are in the Annex and the Article 6(3) exit is shut.
Does the system decide, or materially influence the decision? "The model only recommends, a human approves" is not the defence teams think it is. Where the human follows the recommendation nearly every time, the system is doing the deciding for Annex III purposes.
What attaches if you are in
Chapter III: a risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight design, accuracy and cybersecurity, quality management, conformity assessment, CE marking and EU database registration. Realistically 12 to 18 months of work.
The deadline is 2 December 2027 for Annex III systems, after the June 2026 Omnibus deferral - August 2028 for Annex I products. What the Omnibus did and did not move.
Annex III is amendable by the Commission by delegated act. The list you comply against in 2027 may not be identical to today's. Our changelog records every change that affects classification.
Frequently asked questions
What are the eight Annex III categories?
Biometrics; critical infrastructure; education and vocational training; employment and worker management; access to essential private and public services; law enforcement; migration, asylum and border control; and administration of justice and democratic processes.
Does being in Annex III automatically make my system high-risk?
It makes it presumptively high-risk. Article 6(3) can still remove a system from the category if it performs a narrow procedural task, improves the result of a completed human activity, detects decision patterns without replacing human assessment, or performs a preparatory task - but not if it profiles natural persons.
Is credit scoring in Annex III?
Yes, under point 5, which covers AI evaluating creditworthiness or establishing a credit score. AI used to detect financial fraud is explicitly excluded from that point.
Can the Annex III list change?
Yes. The Commission can amend Annex III by delegated act, so the list you comply against in 2027 may differ from today's.
When do Annex III obligations apply?
2 December 2027, after the June 2026 Omnibus deferred them from August 2026. Annex I systems - AI as a safety component in regulated products - move to August 2028.
Are you inside Annex III?
The free check works through the eight categories and the Article 6(3) exemption in about two minutes.
Start the free scanNo card, no account. Full gap report from €59. See a sample report first. Prices exclude VAT, added at checkout where applicable. 14-day refund.
Self-assessment, not legal advice. This page and the Reglynn report are a structured self-assessment based on the EU AI Act (Regulation 2024/1689, as amended by the June 2026 Omnibus). They indicate where you likely stand and what to verify with a qualified advisor. Reglynn is not a notified body and issues no certification.
How this was made. Reglynn reports are generated by an AI system built on Anthropic's Claude and delivered automatically. Reglynn is a Limited Risk AI system under Article 50; we disclose AI involvement wherever it occurs, including in our chat assistant. Our own self-assessment.