Papers

  • Scaling Hybrid Constrined Zonotopes with optimisation - Winninger T., Urban C., Wei G., Jun 25. Paper

  • Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models - Winninger T., Addad B., Kapusta K., Mar 25. ArXiv / Webpage

Talks

  • Adversarial attacks against reasoning LLMs, Tokyo, NICT, Sep 25.

  • Scaling abstract domains to Large Language Models with Hybrid Constrained Zonotopes, ENS Ulm, INRIA, Jun 25.

  • Mechanistic interpretability for LLM attack and defense, École Polytechnique, CeSIA, Apr 25. Slides

  • Introduction to AI security and reverse engineering, Télécom SudParis, HackademINT, Apr 25. Slides / Webpage

  • Model Poisoning, Station F, CeSIA, Jun 24. Slides

  • GNN based IDS and its robustness against adversarial attacks, Télécom SudParis, HackademINT, Jun 24. Slides

  • Cheating Detection in the 404 CTF, Rendez-vous de la Recherche et de l’Enseignement de la Sécurité des Systèmes d’Information (RESSI), May 24.

  • Introduction to prompt hacking, Télécom SudParis, HackademINT, Nov 23. Slides

  • How to backdoor federated learning, Télécom SudParis, HackademINT, May 23. Slides

  • Introduction to AI & cyber security, Télécom SudParis, HackademINT, May 23. Slides

Research reports

  • Graph Neural Network based Intrusion Detection and its Robustness against Adversarial Attacks, Moreau R., Winninger T., Blanc G., Jun 24. Paper

Posts

  • Subspace Rerouting: Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models, Post

  • Exploring the use of Mechanistic Interpretability to Craft Adversarial Attacks, Post

Hackathons

  • ZaMark: Intellectual Property protection with Homomorphic Watermarking, Privacy Preserving Hackathon, Zama, Sep 24, (finished 2nd). Slides