The four roles
| Role | What it means | Example |
|---|---|---|
| GPAI provider | Builds the underlying AI model | Anthropic (Claude), OpenAI (GPT), Moonshot (Kimi) |
| High-risk AI system provider | Builds or sells an AI system for a high-risk use | A vendor selling an AI credit-scoring system |
| Deployer | Uses an AI system in its own operations | A bank using that credit-scoring system |
| Importer / Distributor | Brings a non-EU AI product into the EU market | A UK firm reselling a US AI product into the EU |
What each owes
GPAI providers
Technical documentation, copyright policy, training-data summary, downstream information (Art. 53).
Applies from 2 Aug 2025
High-risk AI system providers
Risk management, automatic logging, human-oversight design, conformity assessment, post-market monitoring, serious-incident reporting (Arts. 9–15, 19, 43, 72, 73).
Applies from 2 Aug 2026
Deployers
Follow vendor instructions, assign a trained human overseer, monitor operation, retain logs ≥ 6 months, inform affected people, cooperate with regulators (Art. 26).
Applies from 2 Aug 2026
When a deployer becomes a provider
⚠ Watch out — Art. 25
A deployer can become a provider by putting their name on a high-risk system, making substantial modifications to it, or repurposing a general AI model specifically for a high-risk task. When that happens, the full provider obligation stack attaches to the deployer.
Fines
€35M / 7%
of worldwide annual turnover — prohibited practices (Art. 5)
€15M / 3%
of worldwide annual turnover — most other breaches (Art. 99)
Our reading of Regulation (EU) 2024/1689. Not legal advice. Get EU-qualified counsel before acting on this.
· Last reviewed 2026-04-23
· Report a correction