An AI model trained with a large amount of data using self-supervision at scale, that displays significant generality, and can perform a wide range of distinct tasks (Article 3(63)). Examples: GPT-4, Claude, Mistral, Llama.
5 questions answered with specific EU AI Act article references. 98 days until the August 2, 2026 enforcement deadline.
Not sure if your AI system is affected? Take the 5-minute diagnostic.An AI model trained with a large amount of data using self-supervision at scale, that displays significant generality, and can perform a wide range of distinct tasks (Article 3(63)). Examples: GPT-4, Claude, Mistral, Llama.
Technical documentation (Article 53(1)(a)), information for downstream providers (Article 53(1)(b)), copyright compliance policy (Article 53(1)(c)), and training data summary publication (Article 53(1)(d)).
When cumulative training compute exceeds 10^25 FLOPS, or when designated by the European Commission based on capabilities, reach, or known risks (Article 51(2)).
Model evaluation including adversarial testing (Article 55(1)(a)), risk assessment and mitigation (Article 55(1)(b)), incident tracking and reporting (Article 55(1)(c)), and cybersecurity protections (Article 55(1)(d)).
If fine-tuning substantially modifies the model's capabilities, the fine-tuner may become a provider with their own GPAI obligations. Minor fine-tuning for specific tasks generally doesn't trigger new obligations.
All questions and answers in one document. Free.
No spam. Unsubscribe anytime.