AI model evaluation for the European Commission
Selected by the European AI Office to evaluate the manipulation risks of general-purpose AI models, as part of the implementation of the AI Act.
Official evaluator for the AI Act
Analyze
the risks posed by frontier AI
Evaluating the reliability of AI oversight systems
BELLS, notre benchmark pour mesurer la capacité des systèmes d'IA à détecter les comportements problématiques d'autres modèles.
Adopted by the UK AI Safety Institute
Weekly monitoring
Every week, CeSIA analyzes the events shaping the future of AI: technological advances, policy decisions, safety challenges. Independent analysis to understand what's at stake.
Research and analysis
AI safety research, policy recommendations and updates on our work.
Report
A Blueprint for Multinational Advanced AI Development
Dec 2025
Read the report →
Article
Dec 2025
Why an International Agreement Establishing Red Lines for AI is Both Necessary and Realistic
Article
Dec 2025
CeSIA urges European Commission to stay the course on AI Act: "The uncertainty is technological, not regulatory"
Article
Dec 2025
The Global Call for AI Red Lines, Initiated by CeSIA, Is Launched at the UN
Support
institutions in addressing AI challenges
Global Call for AI Red Lines
CeSIA initiated and co-leads an international call for the establishment of red lines in AI development.
As seen in
300+
Personnalités signataires
15
Prix Nobel et Turing
11
Anciens chefs d'État
Bringing the debate to international forums
CeSIA engages in all major international forums and regularly runs workshops and side events to unite AI safety stakeholders.
Highlighted events
Grand Symposium on AI Safety — Sorbonne University
Yoshua Bengio, Stuart Russell and Dragos Tudorache gathered for a dialogue on the safety challenges of the most advanced AI systems.
AI & Media Masterclass — ENS Ulm
With Yoshua Bengio and Reporters Without Borders, 15 leading journalists gathered on AI challenges for the press.
Supporting French institutions in addressing AI risks
From CNIL to the Ministry of Defence, from DGE to the Office of the Minister for Artificial Intelligence: CeSIA supports the full range of institutions shaping French AI policy.
GPAI Code of Practice
Some of our contributions were adopted verbatim in the Code of Practice for general-purpose AI models published by the European AI Office.
Educate
on AI safety issues
Teaching at top universities
Sciences Po, ENS Ulm, ENS Paris-Saclay.
3
cours accrédités dans les grandes écoles
The first comprehensive textbook on AI safety
A comprehensive, regularly updated guide to understanding and mitigating risks from advanced AI systems.
1 000+
étudiants ont suivi un cours avec l'Atlas
International training program
Eight-day in-person bootcamp, with a technical programme and a governance & strategy programme. Originally a CeSIA programme, now run by a partner organization.
200+
formés depuis 2024
98 %
de satisfaction
Get in touch
Working on AI issues and want to connect with our team? Get in touch.