ENS Crypto Team Seminar
Open to the public
To receive the announcements, please subscribe to the ENS Crypto Mailing List
How to find directions
PhD and HdR Defenses
Upcoming crypto events in Paris
Information about cryptography (seminars, conferences, recent results, etc.)
How to subscribe?Amphi Dussane, Defenses
Thales and ENS
PhD Defense of Éric Sageloli.
Jury: Céline Chevalier, Associate Professor, Paris-Panthéon-Assas, University, École normale supérieure, Thesis Supervisor Julia Hesse, Researcher, IBM Research, Thesis Reviewer Dennis Hofheinz, Professor, ETH Zürich, Thesis Reviewer Pierre-Alain Fouque, Professor, University of Rennes, Examiner Dominique Schröder, Professor, TU Wien, Examiner Daniel Rausch, Postdoctoral Researcher, University of Stuttgart, Examiner
Abstract: This thesis contributes to the field of provable security by proposing post-quantum cryptographic constructions and analysing the foundations of universal composability. More specifically, we first present post-quantum constructions based on Euclidean lattices, including DAKE, a family of authenticated key exchange protocols proven secure in strong variants of the extended Canetti-Krawczyk model. These protocols use a double-KEM, a mechanism that works like a KEM but uses two public keys to encapsulate a secret key. We also construct Maul, a double-KEM derived from ML-KEM. When used in DAKE, it yields substantial communication savings over the comparable protocols we study. We then study Canetti's Universal Composability (UC) model, whose composition theorem provides security guarantees for protocols even when executed within a larger system. In our first contribution, we present counterexamples to this theorem, limitations that prevent its application to important classes of protocols, and gaps in its proof. These problems affect nearly all of the model's fundamental notions. We then propose modified versions of the main UC notions to resolve these problems while remaining as faithful as possible to its spirit. The main finding of our second contribution is the redirection problem, linked to how communication between the protocol under study and its environment is handled. This problem has various effects on models whose communication rules are similar to those of Canetti's model. In Canetti's UC model, it reveals that many security statements in the literature are formally false. It also leads to unintended changes in behaviour under certain transformations commonly used in the model and rules out composition in the full generality claimed. Moreover, our attempts to resolve it are unsatisfactory because they come at a substantial cost to simplicity, generality, or modularity. Interface-based models avoid this problem, so it is not inherent to universal composability.
SU
Amphi Jaurès, Crypto Seminar
Tue 10 November 2026, 11:00
Abstract: The rapid development of AI capabilities creates novel risks; international coordination and regulation could be effective to mitigate them. However, given existing distrust and competition between the main actors, low-trust verification mechanisms to check compliance are needed. In this talk, I will give an overview of some of the most promising tools for a working verification regime, including zero-knowledge protocols for inference and training, as well as key research questions.
Joint work with Ky Nguyen (Sorbonne Université, CNRS, LIP6) and Pierre Peigné (General-Purpose AI Policy Lab Paris). Parts of the talk is on preprint https://arxiv.org/pdf/2606.05433