What counts as a high-risk AI system under the EU AI Act?
A system is "high-risk" under the EU AI Act if it falls into one of the two routes in Article 6 of Regulation (EU) 2024/1689, and the special "filter" in Article 6(3) is a narrow exception, not the rule.
How a system becomes "high-risk"
Route 1 — It is a safety component (or is itself a product) covered by EU product-safety law listed in Annex I (Art. 6(1)). This catches AI embedded in toys, lifts, machinery, medical devices, in-vitro diagnostics, motor vehicles, civil aviation equipment, etc. The AI must also be one whose conformity requires an obligatory third-party check by the relevant Union harmonisation act.
Route 2 — It is used in one of the eight sensitive areas listed in Annex III (Art. 6(2)):
| # | Annex III area | Examples |
|---|---|---|
| 1 | Biometrics | Remote real-time identification, emotion recognition at the workplace or in education |
| 2 | Critical infrastructure | Safety components in water, gas, heating, electricity or traffic systems |
| 3 | Education & vocational training | Admission decisions, marking, proctoring |
| 4 | Employment & HR | Recruitment, screening, evaluations, promotion, termination, task allocation |
| 5 | Essential services & private/public credit/insurance | Life-and-health insurance pricing, public-benefit eligibility (with social-security carve-outs), credit scoring |
| 6 | Law enforcement | Risk assessment of offending, polygraphs, evidence reliability, profiling |
| 7 | Migration, asylum, border control | Visa and asylum applications, identity verification |
| 8 | Administration of justice and democratic processes | Judicial decision support, voting systems, election influence |
The "filter" — when Annex III is not high-risk (Art. 6(3))
The Act narrows Annex III down. A system in one of those areas is not high-risk if it performs only:
- a narrow procedural task (e.g. archiving, logging);
- a preparatory task to a human decision (e.g. sorting, pre-screening that does not influence the outcome);
- the detection of decision-making patterns (e.g. fraud flags) where the human reviews the situation and does not rely on it; or
- a preparatory cybersecurity task (e.g. scanning network traffic for incidents).
The provider must document this assessment in line with Article 6(4) and the Commission can challenge it.
Explicit out-of-scope uses (Art. 6(1), second subparagraph; Art. 5)
These are never high-risk (some are restricted; some are simply not covered):
- Systems used only for spam filtering, R&D, or inventory/stock management.
- Standalone AI tools that are not safety components and fall outside Annex III.
- AI used purely for transparency about AI itself, or for non-evidence-based administrative preparation in insurance pricing.
- Biometric verification (1:1 authentication) — only remote identification (1:many) can be high-risk.
- The use of historical statistical facts, learner mastery scoring, or plagiarism checks in education that do not affect the learner's outcome.
Categorically prohibited uses sit in Article 5 (e.g. social scoring, manipulative exploitation, untargeted face-scraping) — these are not "high-risk", they are simply banned.
What changes once the label sticks
A high-risk system must comply with the full Chapter III obligations: risk-management system (Art. 9), data governance (Art. 10), technical documentation and logging (Arts. 11–12), transparency to deployers (Art. 13), human oversight (Art. 14), accuracy/robustness/cybersecurity (Art. 15), plus a CE marking and registration in the EU database (Arts. 43, 49, 71). The Annex III obligations apply from 2 August 2026; Annex III uses that also sit in regulatory sandboxes under Article 53 are exempt until 2 August 2028.
This article is general information about Regulation (EU) 2024/1689, not legal advice. The exact classification can depend on factual and technical details, so providers and deployers should document their Article 6(2)–(3) assessment and check the implementing acts and Commission guidance.
Sources: Regulation (EU) 2024/1689 (EU AI Act) — Art. 5, Art. 6, Art. 7, Art. 9–15, Art. 43, Art. 49, Art. 53, Art. 71, Annex I, Annex III.