EU AI Act High-Risk AI Systems

High-risk AI systems are AI applications that must meet particularly strict requirements under Article 6 of the AI Act. There are two categories: First, AI systems integrated as safety components in products under Annex I (e.g. medical devices, machinery, lifts). Second, standalone AI systems in the eight application areas of Annex III — from biometric identification through employment management to law enforcement. Importantly, AI systems in Annex III areas can still be classified as non-high-risk if they have no significant impact on decision-making (Art. 6(3)).

Under Article 16 of the AI Act, providers must fulfil extensive obligations: Establish a risk management system (Art. 9), ensure data governance (Art. 10), create and maintain technical documentation (Art. 11), implement automatic logging (Art. 12), ensure transparency and user information (Art. 13), enable human oversight (Art. 14), ensure accuracy, robustness, and cybersecurity (Art. 15), operate a quality management system (Art. 17), and conduct a conformity assessment with CE marking (Art. 43).

Annex III defines eight high-risk areas: 1) Biometric identification and categorisation. 2) Management and operation of critical infrastructure (transport, energy, water, digital networks). 3) Education and vocational training (exam assessment, access control). 4) Employment and worker management (recruitment, promotion, dismissal). 5) Access to essential public and private services (creditworthiness, emergency services, insurance). 6) Law enforcement. 7) Migration, asylum, and border control. 8) Administration of justice and democratic processes.

Before placing on the market, every high-risk AI system must undergo a conformity assessment (Art. 43). For most Annex III systems, this is conducted as an internal control by the provider (Annex VI). For real-time remote biometric identification in law enforcement, a third-party assessment by a notified body is required (Annex VII). The provider then issues an EU declaration of conformity (Art. 47), affixes CE marking (Art. 48), and registers the system in the EU database. A renewed assessment is required for substantial modifications.

Article 14 requires that high-risk AI systems be designed to allow effective oversight by natural persons. The oversight must enable: understanding of system capabilities and limitations, detection and remediation of anomalies and unexpected outputs, the ability to disregard the system or intervene in the output, and system interruption via a stop function. The oversight measures must be proportionate to the risk level and degree of autonomy of the system.

Obligations take effect in stages: Since 2 February 2025, prohibitions on unacceptable-risk AI apply (Art. 5). Since 2 August 2025, GPAI rules apply. The Digital Omnibus on AI, adopted in June 2026, moved the high-risk requirements under Annex III from 2 August 2026 to 2 December 2027. For Annex I high-risk systems (product safety), the deadline is now 2 August 2028. Systems already on the market must be updated when substantially modified.

Article 72 obliges providers to establish a post-market monitoring system that actively and systematically collects and analyses performance data throughout the AI system’s entire lifecycle. This includes user complaints, malfunctions, and incidents. For serious incidents — those that have caused or may cause death, serious health damage, or significant property damage — Article 73 mandates reporting to the market surveillance authority within 15 days. The monitoring must be set out in a documented plan and updated regularly.