About me
I work at the intersection of cybersecurity and AI safety.
In January 2025, I’ll be starting a PhD at INRIA, in their new team focused on AI security and auditing: ARTISHAU, in Rennes (France).
Some things I’ve done in the past:
- 2024: published eyeballvul: a future-proof benchmark for vulnerability detection in the wild.
- 2023-2024: designed tasks to evaluate dangerous capabilities of AI models (especially in offensive cybersecurity), as a contractor for METR.
- summer and autumn 2023: attended MATS, where I worked on evaluations of autonomous hacking capabilities in AI models (writeup here).
- 2021-2022: worked at Red Balloon Security for a year and a half, as a vulnerability researcher (we discovered CVE-2022-22723 and CVE-2022-22725 in Schneider P3 and P5 protection relays, and showed for the first time that embedded devices could be hit by ransomware), and as a software engineer.
- 2017-2021: studied at École Polytechnique (X2017)
- 2017: worked full-time as a teacher in prison for 6 months