EU Artificial Intelligence Act (Regulation 2024/1689)

Summary

The EU AI Act is the world's first comprehensive legal framework for artificial intelligence, establishing risk-based regulations for AI systems deployed in the European Union. It categorizes AI into unacceptable risk, high-risk, limited risk, and minimal risk tiers, with corresponding obligations for transparency, safety, and conformity assessment. It also introduces specific rules for general-purpose AI models, including those with systemic risk.

Key Points

• •Establishes a four-tier risk classification: prohibited AI practices, high-risk AI systems, limited-risk AI, and minimal-risk AI with proportionate obligations for each.
• •Imposes strict requirements on high-risk AI systems (e.g., biometrics, critical infrastructure, employment) including transparency, human oversight, and conformity assessments.
• •Introduces rules for general-purpose AI (GPAI) models, with enhanced obligations for those deemed to pose systemic risk (e.g., models trained with >10^25 FLOPs).
• •Establishes the European AI Office to oversee enforcement, particularly for GPAI models, alongside national competent authorities.
• •Prohibits certain AI applications outright, including real-time remote biometric identification in public spaces and social scoring by public authorities.