← All frameworks
Binding

EU AI Act

European Commission / AI Office · Brussels, EU · effective 1 Aug 2024 (phased)

The first comprehensive, binding AI law. Risk-tiered (unacceptable / high / limited / minimal), with GPAI model rules and heavy transparency duties. Extraterritorial — applies to any provider putting AI on the EU market.

Who must comply
  • Providers & deployers of AI in the EU
  • GPAI model providers globally
  • Importers, distributors, product manufacturers
Penalties

Up to €35M or 7% of global annual turnover for prohibited-practice breaches.

Key obligations
Risk classification
Classify every system: prohibited, high-risk (Annex III), limited or minimal.
High-risk conformity
Risk management, data governance, logging, human oversight, accuracy & robustness, technical documentation.
Transparency (Art. 50)
Disclose AI interaction, label deepfakes and AI-generated content.
GPAI duties
Model documentation, copyright policy, systemic-risk evals for frontier models.
Sectors in scope
HealthcareFinanceHR / hiringCritical infrastructureLaw enforcementEducation
Threats & cybersecurity it addresses
Biometric mass surveillanceUnsafe high-risk automationUndisclosed deepfakesFrontier / systemic model risk
Crosswalks — comply once, cover many
CSOAI Layer 0 mapping

Charter Art. 2, 17, 24, 33, 39

Governed MCP tools — open source, pip/npx install