Centre pour la Sécurité de l'IA (CeSIA)

The Centre pour la Sécurité de l'IA (CeSIA), known in English as the French Center for AI Safety, is an independent French nonprofit (association loi 1901) headquartered in Paris, dedicated to preventing and mitigating major AI risks through policy research, public education, and direct engagement with policymakers. Founded in 2024 by a team emerging from EffiSciences — a French AI safety field-building organization — CeSIA brought together academics, policy professionals, and civil society practitioners to create France's first dedicated AI safety center. It accepts no funding from private AI companies.

Founding and organization

CeSIA was established in 2024 by Charbel-Raphaël Ségerie (Executive Director), Florent Berthet (Director of Operations), Manuel Bimich (Head of Strategy), and Alexandre Variengien, with Vincent Corruble — an associate professor at Sorbonne University — serving as Scientific Director. The organization grew out of EffiSciences' two years of AI safety field-building work in France, which had already produced the ML4Good bootcamp program and accredited university courses. By 2025–2026 the team had grown to include dedicated staff across policy engagement, EU advocacy, technical governance, and communications — with a Senior EU Engagement Officer based in Brussels and a Senior Policy Analyst alongside the Paris-based core team.

Funding comes from the European Commission, ENS (École Normale Supérieure), the UK AI Security Institute, and European and British philanthropic organizations including Longview Philanthropy, the Survival and Flourishing Fund, the AI Safety Tactical Opportunities Fund, and Effektiv Spenden. CeSIA's partnership network lists the European Commission, UK AISI, OECD, Sciences Po, ENS Ulm, and UNESCO as institutional collaborators.

Programs and activities

CeSIA works across four complementary pillars: academic field-building, AI governance, technical R&D, and public outreach.

Academic field-building. CeSIA offers the first accredited AI safety courses in France, running at ENS Ulm and the MVA master program. The ML4Good bootcamp — originally developed through EffiSciences — has been replicated over ten times globally, seeding AI safety community capacity in multiple countries. CeSIA also operates a mentoring program for researchers and practitioners entering the field.

AI governance and policy. CeSIA engages directly with French and European policymakers. Its consultation was cited eleven times in the AI Action Summit expert report and contributed to drafting the EU AI Act Code of Practice. The organization has advised the French Senate, the Ministry of Defense, and the French national AI body INESIA on frontier AI risks. At the international level, it co-authored reports including "Governing the Frontier: Priorities for International Cooperation on Advanced AI" (November 2025) and "A Blueprint for Multinational Advanced AI Development" (December 2025), and contributes to the OECD's AI governance frameworks.

Technical R&D. CeSIA developed BELLS (Benchmark for the Evaluation of LLM Safeguards), published at ICML. Since 2025 it has been contracted by the European Commission to evaluate frontier AI models for manipulation risks — one of the first civil-society organizations to carry out frontier-model evaluation under European regulatory authority.

Public outreach. CeSIA collaborated with French-language YouTubers to produce AI safety content reaching approximately three million views, and has co-authored op-eds with researchers including Yoshua Bengio. It publishes a weekly AI safety policy analysis newsletter and runs public events across Europe, including workshops co-hosted with Sciences Po and UNESCO.

Global Call for AI Red Lines

CeSIA's highest-profile work to date is initiating the Global Call for AI Red Lines, an international civil society declaration launched on 22 September 2025 at the 80th UN General Assembly high-level week in New York. CeSIA co-organized the initiative with The Future Society and UC Berkeley's Center for Human-Compatible AI (CHAI). Yoshua Bengio presented the declaration at a concurrent UN Security Council session on AI, following Secretary-General António Guterres's call to action against lethal autonomous weapons. Nobel Peace Prize laureate Maria Ressa announced the initiative during the UNGA opening session. By September 2026 the declaration had accumulated 746+ individual signatories and endorsement from 90+ organizations. CeSIA co-organized a six-workshop series across five countries — Athens, Delhi, Singapore, and twice in Paris at UNESCO and Sciences Po — within the first year of launch. The campaign is tracked in the corpus under camp-global-call-for-ai-red-lines.