General Purpose AI Policy Lab

The GPAI Policy Lab (General-Purpose AI Policy Lab) is an independent French nonprofit association formally registered on February 21, 2025, and headquartered at Campus Cyber in La Défense, Paris. It was incubated at Sciences Po Paris and operates under the motto "Understand, Coordinate, Act." The organization's core mission is to work alongside French and European institutions to address the security and international coordination requirements posed by the development of general-purpose AI (GPAI). The lab aims to serve as a bridge between cutting-edge technical expertise in AI safety and the operational needs of French security, defense, and policy institutions. The GPAI Policy Lab produces a series of concise AI Security Policy Briefs that cover foundational topics in the AI safety and security literature, including the limits of current interpretability techniques, situational awareness and evaluation awareness, reward hacking, model agency, and autonomous replication. These briefs are tailored for stakeholders in French governmental and institutional settings. Beyond research outputs, the lab monitors frontier AI capabilities developed in the US and China — particularly in domains such as cyberhacking and persuasion — and provides strategic advisory services to French institutional actors. It also runs training programs, including a six-week intensive training program at Sciences Po that brings together students, early-career professionals, and leading AI governance experts. The organization is led by Tom David (President), a graduate of the École Normale Supérieure de Lyon and co-founder of PRISM Eval, a firm specializing in stress-testing generative AI models on critical behaviors. Other core team members include Pierre Peigné, Lola Carbonell, Benoît Larher, and Jérémy Andréoletti, an AI Security Researcher with a background in computational evolutionary biology. As of 2025, the organization had approximately nine employees.