The EU Commission has published guidelines for GPAI models establishing a computational threshold of 10²³ FLOPs as a classification criterion; models exceeding 10²⁵ FLOPs are considered high-risk and subject to extensive reporting, documentation, and risk assessment requirements, with providers able to contest classifi

Modifiers of AI systems or GPAI models must assess whether their modifications result in provider classification — misclassification risks fines up to €15 million or 3% of global annual turnover.

Article 50 mandates AI transparency disclosure in four scenarios: direct AI-human interaction, synthetic content, emotion and biometric recognition, and deepfakes and AI-generated text on matters of public interest.