AI Risk Categories

aka AI Act risk tiers, AI risk classification

Four-tier risk classification under the EU AI Act: unacceptable-risk (banned), high-risk (regulated), limited-risk (transparency), minimal-risk (unregulated). Determines a system's compliance burden.

Last reviewed April 2026

Definition

The EU AI Act classifies AI systems and models into four risk tiers, and each tier dictates the applicable obligations. Unacceptable-risk practices (Article 5) are prohibited outright: cognitive-behavioural manipulation, social scoring by public authorities, exploitation of vulnerable groups, untargeted scraping of facial images for biometric databases, real-time remote biometric ID in public spaces (with narrow law-enforcement exceptions), emotion recognition in workplaces or schools, and biometric categorisation by sensitive attributes. High-risk AI systems (Annex III) include AI used in employment (CV screening, performance management), credit scoring, education access, critical infrastructure, biometric identification, justice administration, and migration/border control, which face full conformity assessment, technical documentation, human oversight, post-market monitoring, and CE-marking obligations on providers, plus deployer obligations on the businesses that use them. Limited-risk AI (chatbots, deepfakes, AI-generated content) faces transparency obligations: users must be informed they are interacting with AI or viewing AI-generated content. Minimal-risk AI (spam filters, AI in video games, recommender systems below sensitivity thresholds) is unregulated by the Act. General-Purpose AI (GPAI) models sit in a parallel framework with their own obligations, escalating sharply once a model crosses the systemic-risk threshold of 10^25 FLOPs of training compute.

Why it matters for software choice

Most software marketed as 'AI-powered' is minimal-risk or limited-risk and triggers no Act obligations on the buyer. But anything used to make decisions about hiring, credit, education access or essential services likely crosses the high-risk line, and the buyer (deployer) shares the compliance burden with the vendor. Verify the risk tier in writing before adoption.

Authority sources

Software categories this affects

Vendors covered by this term

Related terms