EU-AI-ACT-FOUNDATION-MODELS

EU AI Act Foundation Models AI safety + governance framework.

Définition

EU AI Act Foundation Models / GPAI obligations designe les specific obligations pour General-Purpose AI Models (GPAI, foundation models like GPT-4 + Claude + Gemini + Llama + Mistral) sous EU AI Act 2024/1689, distinguish standard GPAI (Article 53) + GPAI Models with Systemic Risk (Article 55, computational threshold 10^25 FLOPs training), obligations Article 53 (technical documentation + copyright compliance + summary training data + transparency + cooperation AI Office), Article 55 additional (model evaluation + systemic risk assessment + serious incident reporting + cybersecurity). Framework + standard + guidance objectifs AI safety + governance + risk management + transparency + fairness + accountability + explainability + multiple AI ethics principles adoption multi-stakeholder process + voluntary adoption industry + emerging regulation alignment EU AI Act + US Executive Orders + multiple national jurisdictions + private sector adoption Microsoft + Google + Meta + OpenAI + Anthropic + autres major AI labs + corporate compliance teams 2020s+.

Origine

EU AI Act 2024/1689 GPAI provisions effective 2 aout 2025 obligations + GPAI Code of Practice ; Systemic Risk threshold 10^25 FLOPs training.

Exemple en contexte

OpenAI commercialise GPT-5 in EU : if training compute exceeds 10^25 FLOPs threshold, classified GPAI Model with Systemic Risk per Article 55 ; OpenAI must conduct model evaluations standardized protocols + assess + mitigate systemic risks (bias + cybersecurity + societal harms + democratic integrity + critical infrastructure risks) + report serious incidents AI Office + maintain adequate cybersecurity practices.

Termes liés

• AI Act EU detail — parent regulation.