← All entries

AI regulation glossary

AI system

A machine-based system that, with some autonomy, infers from input how to generate outputs — predictions, recommendations or decisions — that influence physical or digital environments.

Provider

Whoever develops an AI system (or has it developed) and places it on the market under their own name. Carries the heaviest obligations.

Deployer

Whoever uses an AI system under their own authority (not personal use). Has their own, lighter duties than the provider.

GPAI (general-purpose AI)

A broadly usable AI model, such as a large language or image model. Has its own regime (Chapter V).

Systemic risk

A particular risk of the most capable GPAI models; presumed above 10^25 FLOP of training compute. Triggers extra duties (Art. 55).

High risk

AI in sensitive uses (Annex III) or as a safety component in regulated products (Annex I). The heaviest set of obligations.

Prohibited practice

AI use that is banned outright (Art. 5), such as social scoring or manipulation. In force since 2 February 2025.

Conformity assessment

The procedure by which a provider demonstrates, before market entry, that a high-risk system meets the requirements.

FRIA (fundamental-rights impact assessment)

Certain deployers assess, before use, a high-risk system's impact on fundamental rights (Art. 27).

Transparency obligation

The duty to make AI recognisable (Art. 50): disclose chatbots, mark synthetic output, label deepfakes.

AI literacy

A sufficient level of AI knowledge among the staff who work with it (Art. 4). In force since 2 February 2025.

AI Office

The European AI Office supervises GPAI models and coordinates implementation of the AI Act.

Digital Omnibus

A simplification package (political agreement 7 May 2026) that, among other things, shifts the application dates of high risk and transparency.

FLOP

Floating-point operation, a unit of compute. The EU systemic-risk threshold sits at 10^25 FLOP of cumulative training compute.