The French Center for AI Safety (Centre pour la Securite de l'IA) is a Paris-based non-profit think tank and research center working to reduce risks from artificial intelligence through education, technical research, and policy advocacy in France and Europe.
The French Center for AI Safety (Centre pour la Securite de l'IA) is a Paris-based non-profit think tank and research center working to reduce risks from artificial intelligence through education, technical research, and policy advocacy in France and Europe.
People
Updated 05/18/26Executive Director and Co-Founder
Head of Advocacy
Head of Public Policy / Head of Policy
Director of Operations
Fellow
Head of Technical Governance
Head of Policy Engagement
Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26CeSIA (Centre pour la Securite de l'IA), the French Center for AI Safety, is an independent non-profit organization based in Paris that works to build a culture of AI safety in France and Europe. Launched on May 31, 2024, CeSIA evolved from the AI division of EffiSciences, a student-led French organization focused on existential risk research. The organization's work spans four strategic pillars: academic training, research and development, institutional advocacy, and public awareness. In education, CeSIA provides Europe's first accredited university-level AI safety courses at ENS Ulm and Paris-Saclay University through the Turing Seminar, and runs the international ML4Good bootcamps, which have trained over 500 students and professionals across eight countries including France, Germany, the UK, Brazil, Singapore, Colombia, Italy, and Switzerland. CeSIA is also developing a MOOC with ENS Paris-Saclay and publishes the AI Safety Atlas, an open-source textbook covering AI capabilities, risks, alignment, and governance. On the technical research side, CeSIA develops the BELLS benchmark, which evaluates the robustness of safety monitoring systems for large language models, establishing one of the first frameworks for third-party evaluation of AI safeguards. The team also conducts research on scalable supervision systems for LLM-based agents and warning shot theory. In advocacy, CeSIA co-led the Global Call for AI Red Lines with The Future Society and CHAI, which was endorsed by over 300 personalities including 12 Nobel laureates and presented at the 80th United Nations General Assembly in September 2025. The organization contributes to EU AI Act Code of Practice development, works with the OECD on AI safety, and provides advisory support for French AI evaluation efforts. CeSIA was co-founded by Charbel-Raphael Segerie (Executive Director), Florent Berthet (Operations Director), Manuel Bimich (Strategy Director), and several other team members. The organization is supported by European and British philanthropic organizations including Longview Philanthropy, Effektiv Spenden, the Survival and Flourishing Fund, and AISTOF, and does not accept funding from private AI companies to avoid conflicts of interest.
Theory of Change
Updated 05/18/26CeSIA believes that reducing catastrophic AI risks requires building a robust culture of AI safety in France and Europe, regions that play a decisive role in global AI governance. Their theory of change operates through three channels: (1) training the next generation of AI safety researchers and engineers through university courses, bootcamps, and open educational resources, thereby growing the field; (2) developing technical evaluation tools like the BELLS benchmark that enable third-party assessment of AI safeguards, creating accountability mechanisms for AI developers; and (3) translating technical AI safety concerns into actionable policy recommendations for European and international institutions, helping establish regulatory frameworks and red lines that prevent the most dangerous AI capabilities from being deployed without adequate safeguards.
Grants Received
Updated 05/18/26Projects
Updated 05/18/26
An open, comprehensive online textbook and learning ecosystem on AI safety—covering AI capabilities, risks, alignment, and governance—authored by researchers from the French Center for AI Safety (CeSIA).
An open-source benchmark suite developed by CeSIA to evaluate and compare large language model supervision and safeguard systems, measuring how reliably they detect problematic or unsafe behaviour in other models.
An international campaign co-initiated and co-led by CeSIA calling on governments to agree binding ‘red lines’ that prohibit universally unacceptable AI uses and capabilities, launched around the 80th UN General Assembly and endorsed by hundreds of prominent figures and organisations.
Discussion
Key risk: As a small, early-stage org stretching across education, technical benchmarking, and international advocacy, their theory of change relies on EU AI Act enforcement and ‘red lines’ becoming meaningfully adopted—both uncertain—so execution capacity and counterfactual impact may be limited.
Case for funding: CeSIA sits at the nexus of technical evaluation and EU policy, with demonstrated traction (their recommendations were adopted verbatim into the EU AI Act’s Code of Practice) and unique European talent-pipeline assets (the only accredited university AI safety courses and the AI Safety Atlas), positioning them to make EU enforcement and third‑party evaluations like BELLS actually bite.