High-risk AI system

In plain English

Think of airport security. Most passengers pass through light checks, a few items are banned outright, and some cargo needs special handling. High-risk is the special-handling lane: allowed, but only with testing, paperwork and supervision.

Status depends on the use, not the technology. The same model can be minimal-risk in a playlist app and high-risk when it ranks job applicants.

Example

Orrin Systems, a fictional HR-tech company, sells two modules to EU employers.

• Module A converts CVs into a standard format and files them by job. It scores nobody. Orrin documents that it performs only a narrow procedural task under Art. 6(3), registers it, and treats it as not high-risk.
• Module B ranks candidates by predicted job fit. That is an Annex III employment use (point 4), and it profiles people, so the Art. 6(3) exit is closed. It is high-risk.

For Module B, Orrin as provider must meet Arts. 9 to 15 and run a quality management system. Before market placement, it must pass a conformity assessment (internal control for employment uses) and sign an EU declaration of conformity. It must then affix the CE marking and register in the EU database. An employer using Module B becomes a deployer with Art. 26 duties. These include assigning trained human oversight, keeping logs for at least six months and informing workers before use.

Why it matters

The EU AI Act has two doors into high-risk and one narrow exit (Art. 6):

• Door 1, Art. 6(1) and Annex I: AI that is a regulated product, or a safety component of one, such as a medical device. The product must need third-party conformity assessment. The 2026 Omnibus clarified that AI used only for non-safety tasks, such as user assistance or performance optimization, is not a safety component.
• Door 2, Art. 6(2) and Annex III: eight areas of use. They are biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration and borders, and justice and democracy. Essential services include public benefits, credit scoring, life and health insurance pricing and emergency triage.
• The exit, Art. 6(3): an Annex III system is not high-risk if it poses no significant risk of harm. Qualifying tasks are narrow procedural tasks, improving a completed human activity, detecting patterns without replacing human assessment, and preparatory work. A system that profiles people is always high-risk. The provider must document its assessment (Art. 6(4)) and register (Art. 49(2)).

Dates. The Digital Omnibus moved high-risk obligations to 2 Dec 2027 (Annex III) and 2 Aug 2028 (Annex I).

Penalties. Breaching high-risk obligations can cost up to €15 million or 3% of worldwide annual turnover, whichever is higher (Art. 99(4)). For SMEs, and since the Omnibus small mid-caps, the cap is whichever is lower.

How the AIGP exam tests it

Classification is the core skill (lesson II.C.3), followed by provider duties (II.C.4, II.C.5).

Common traps:

• "A human reviews it, so it isn't high-risk." Human review never changes the classification.
• Annex III near misses. Fraud detection is carved out of the credit point. Car insurance pricing is not listed (only life and health). Biometric verification such as face unlock is excluded.
• The wrong door. A diagnostic medical device enters through Annex I. Health care as such is not an Annex III area.
• Old dates. Answers built on 2 Aug 2026 for Annex III are out of date.

The tell: ask what the system is used for, then check whether it profiles people.