Centre pour la Sécurité de l'IA
Budget
Not declared
EP Access
0
accredited persons
Staff
10
7.5 FTE
EU Grants
None
Mission & Goals
CeSIA is a Paris-based non-profit organization dedicated to understanding, preventing, and mitigating the risks associated with artificial intelligence. Our core mission is to foster an AI safety culture within France and Europe by promoting technical, policy, and regulatory approaches to address critical AI challenges.
EU Legislative Interests
We initially registered in the transparency register in order to participate in the drawing-up of the first General-Purpose AI Code of Practice. Now we keep this information updated in order to apply to potential tenders from the EU commission, notably.
Communication Activities
The Centre pour la Sécurité de l’IA (CeSIA) is a center for AI safety in Paris that focuses on the safety of general-purpose AI systems. We have three missions: 1) Informing the public and policy discourse around AI safety: - We organized a series of round tables with many important actors (OECD, Mistral, Mozilla, CNIL, Mila, LNE, etc), discussing many topics related to the General-Purpose AI (GPAI) Code of Practice. - Partnered with the OECD to share information on risk thresholds for GPAI. - Published an Op-Ed with Yoshua Bengio: https://www.securite-ia.fr/post/english-version-securite-de-lia-il-est-urgent-de-definir-des-lignes-rouges-a-ne-pas-franchir , to explain why we need thresholds of capabilities for GPAI 2) R&D: - Published an AI evaluation benchmark at the ICML NextGen Workshop for the evaluation of LLM supervision, in partnership with GiskardAI and Sorbonne: https://www.securite-ia.fr/en/post/bells-un-benchmark-pour-evaluer-la-fiabilite-des-systemes-de-supervisions-des-llms - Creating a textbook on AI Safety for GPAI: https://ai-safety-atlas.com/ 3) Academic teaching and helping talents pursue AI safety research: - We offer the only university courses on AI safety for general-purpose AI systems in France, at ENS Ulm and ENS Paris-Saclay: https://www.master-mva.com/cours/seminaire-turing/ - Since 2022, we have organized over 10 research bootcamps on AI safety in Europe, recently funded by Erasmus+: https://www.ml4good.org/
Interests Represented
Does not represent commercial interests
Member Of
The CeSIA is part of the European Network for AI Safety: https://www.enais.co/network
Organisation Members
The list of members can be found here: https://www.securite-ia.fr/a-propos
Additional Information
CeSIA has been incubated by EffiSciences, a French non-profit.
Commissioner Meetings
No recorded meetings with EU commissioners.