EU AI Act glossary

High-risk AI system

A high-risk AI system is one used for a purpose the EU AI Act lists, such as screening job applicants, marking students or checking credit scores, or a safety part of certain regulated products.

In plain English

High-risk systems carry the AI Act's heaviest obligations: risk management, data governance, technical documentation, logging, human oversight, accuracy and security requirements, a conformity assessment and, for most systems on the Annex III list, registration in an EU database. These fall mainly on the provider, but whoever uses the system has duties too, such as human oversight and, for public bodies, public services, credit scoring or life and health insurance pricing, a fundamental rights impact assessment.

There are two routes into the category. The first is AI that is a safety component of a product covered by EU product-safety law, or is itself such a product, where the product needs an independent safety check (for example a medical device, toy or lift). The second, more relevant to most builders, is AI used for one of the purposes listed in Annex III. These are: biometrics (identifying people from their features, sorting people by sensitive traits using their biometric data, and recognising emotions, but not simple identity checks such as a face login); safety parts of critical infrastructure; education (admissions, marking, exam monitoring); employment (job-ad targeting, CV screening, hiring decisions, performance monitoring); checking individuals' creditworthiness (fraud detection excluded); pricing individuals' life and health insurance; public benefits decided by or for public bodies; assessing emergency calls, dispatching emergency services and emergency triage; and policing, migration and border control, helping courts or arbitration bodies research and apply the law, and influencing how people vote.

A system on the Annex III list is not high-risk if it does not pose a significant risk, for example because it only does a narrow procedural or preparatory task without materially affecting decisions. That exception never applies to systems that profile people. A provider who relies on it must document why before launch, still register the system, and show the assessment to regulators if asked.

The Digital Omnibus on AI moved the start dates: 02/12/2027 for the Annex III areas and 02/08/2028 for AI in regulated products.

What it means for your build

  • Building anything that screens job applicants, marks or admits students, checks individuals' creditworthiness, or prices life or health insurance? Stop and get specialist advice before going further.
  • Watch the purpose, not just the tool. From 02/12/2027, a client who uses a general chatbot to screen CVs is likely to turn it into a high-risk system and become its provider (Article 25).
  • Some uses have been banned since 02/02/2025, not just treated as high-risk: inferring emotions at work (including recruitment) or in education, except for medical or safety reasons, and using biometrics to work out someone's race, politics, religion, trade union membership, sex life or sexual orientation.
  • Even before 02/12/2027, data protection law (UK and EU GDPR) already applies to these systems if they use personal data, including the rules on significant decisions made solely by automated means.
  • If you rely on the "no significant risk" exception, write down your reasoning before launch. The provider has to keep that assessment and still register the system.

Sources

Related terms

Which of these apply to your build?

Answer about a dozen plain-English questions and get a checklist for the website, AI agent or automation you are building.

Check your build

Guidance, not legal advice. Last checked 13/11/2026. ยท How we check