I am a Scientific Lead in the AI Department at CNIL, the French Data Protection Authority, where I work on AI safety, privacy and the technical auditing of AI systems in the context of European regulation.
My academic background is in Topological Data Analysis (Ph.D., INRIA/École polytechnique; postdoc at EPFL). My publications are on the Research page; my technical projects and articles are on the Projects page.
Current focus
-
PANAME — project manager of a partnership of 30+ researchers and developers across four institutions, building a unified open-source Python library for GDPR compliance and privacy-preserving AI evaluation. About PANAME -
AI safety & privacy research — research roadmaps on model quantization and differential privacy (with INRIA) and on the sociotechnical dimension of explainable and open-source AI (with Sciences Po); representing CNIL in PEPR Cybersécurité and at INESIA, the French AI Safety Institute. -
AI Act auditing & technical investigations — design of CNIL's technical auditing framework for AI systems under the EU AI Act, and leading robustness investigations on algorithmic and generative AI systems (vulnerabilities, bias, data leakage). -
Policy contribution — co-author of national guidelines (CNIL's AI doctrine) and EDPB frameworks on AI robustness, security and model confidentiality. CNIL's practical AI guides
News
-
New open-source tool: genmod, an interactive genealogy of open-source AI models, is now available on Hugging Face. It grows out of our LINC work on open-source AI with Anna Charles: read the project presentation.
-
PANAME — the partnership between CNIL and ANSSI with academic institutions for auditing the confidentiality of AI models — is rolling out: see the announcement on cnil.fr and the press coverage.
-
Embrace the opaque: A sociology of the xAI community (with Mehdi Arfaoui, Romain Pialat, Pierre François and Laurence Barry) — a socio-technical study of the explainable-AI research field — has been published.
-
The technical report Techniques d'audit des grands modèles de langage (with Henri Duprieu) — evaluation frameworks and safety auditing guidelines for large language models — is now available.
Recent events
- December 2025 — CNIL LINC seminar: presentation of the project experimenting with rights-exercise scenarios to inform CNIL positions, with Anna Charles.
- July 2025 — Invited speaker at the first INESIA Scientific Days, Centre Inria de Paris, France.
- March 2025 — Invited speaker at the ELSA Workshop on Privacy Preserving Machine Learning. Privacy of AI systems: the point of view of a Data Protection Authority, Bertinoro, Italy.
- July 2024 — Contributed talk at the EASST'24 workshop on the sociotechnical dimension of explainable and transparent AI. The field of explainable AI: machines to explain machines? Amsterdam, Netherlands.
- May 2024 — Invited speaker at the ICLR'24 workshop on Privacy Regulation and Protection in Machine Learning. Technical and scientific stakes of ML regulation: a Data Protection Authority perspective, Vienna, Austria.