Article 51

Article 51 of Regulation (EU) 2024/1689 establishes the classification of general-purpose AI models with systemic risk. The main presumptive criterion is the amount of FLOPs used in training: above 10²⁵ FLOPs, systemic risk is presumed. Today this threshold captures GPT-4, Claude Opus, Gemini Ultra, and equivalents. It is the gateway to the substantive regime of Article 55. Appears in The Faking Machine in the contextualisation of the GPAI tier and of the architecture of supplementary obligations the regulation has built on top of frontier models.