The European Artificial Intelligence Regulation
The EU AI Act (Regulation 2024/1689) is the world's first comprehensive AI legislation. It classifies AI systems by risk levels.
Risk Classification
- Unacceptable risk (prohibited): subliminal manipulation, social scoring, real-time remote biometric identification in public spaces
- High risk (Annex III): includes systems used in the administration of justice
- Limited risk: chatbots (transparency obligation)
- Minimal risk: most AI applications
Annex III: Justice as High Risk
Annex III expressly includes AI systems used to:
- Assist judicial authorities in investigating facts and law
- Apply law to specific cases
- Influence election or referendum outcomes
This means any AI tool used by a court to assist in sentencing, analyze case law, or assess procedural risks falls under the high risk category.
Obligations for High-Risk Systems
- Documented risk management system
- High-quality, representative training data
- Logging and traceability of decisions
- Transparency: clear information about how it works
- Effective human oversight
- Accuracy, robustness, and cybersecurity
Application Timeline
- February 2025: unacceptable risk prohibitions
- August 2025: obligations for general-purpose models
- August 2026: full application of Annex III (high risk)