Attaques et preuves de sécurité des protocoles d'échange de clés authentifiés

The vast majority of communication on the Internet and private networks heavily relies on Public-key infrastructure (PKI). One possible solution, to avoid complexities around PKI, is to use Password Authenticated Key-Exchange (PAKE) protocols. PAKE protocols enable a secure communication link between the two parties who only share a low-entropy secret (password).
PAKEs were introduced in the 1990s, and with the introduction of the first security models and security proofs in the early 2000s, it was clear that PAKEs have a potential for wide deployment - filling the gap where PKI falls short. PAKEs' PKI-free nature, resistance to phishing attacks and forward secrecy are just some of the properties that make them interesting and important to study. This dissertation includes three works on various aspects of PAKEs: an attack on an existing PAKE proposal, an application of PAKEs in login (for password leak detection) and authentication protocols (HoneyPAKEs), and a security analysis of the J-PAKE protocol, that is used in practice, and its variants. In our first work, we provide an empirical analysis of the zkPAKE protocol proposed in 2015. Our findings show that zkPAKE is not safe against offline dictionary attacks, which is one of the basic security requirements of the PAKE protocols. Further, we demonstrate an implementation of an efficient offline dictionary attack, which emphasizes that, it is necessary to provide a rigorous security proof when proposing a new protocol. In our second contribution, we propose a combined security mechanism called HoneyPAKE. The HoneyPAKE construction aims to detect the loss of password files and ensures that PAKE intrinsically protects that password. This makes the PAKE part of the HoneyPAKE more resilient to server-compromise and pre-computation attacks which are a serious security threat in a client-server communication.
Our third contribution facilitates the wider adoption of PAKEs. In this work, we revisit J-PAKE and simplify it by removing a non-interactive zero knowledge proof from the last round of the protocol and derive a lighter and more efficient version called sJ-PAKE. Furthermore, we prove sJ-PAKE secure in the indistinguishability game-based model, the so-called Real-or-Random, also satisfying the notion of perfect forward secrecy.