Protecting security protocols against guessing attacks: towards

Designing secure protocols based on passwords is a difficult task. Indeed, passwords, and more generally low-entropy secrets, are potentially vulnerable to guessing attacks, that is, exhaustive, "brute force" searches. Preventing guessing attacks typically requires a protocol to conceal any partial information on the password (e.g. its checksum) which could help the attacker confirm his guess during the searching process.<br/> Ensuring such a property is not obvious. Based on the seminal work of Lowe, several models and automatic tools have been proposed to analyze protocols with respect to guessing attacks. Unfortunately, these works rely on different symbolic models for which no computational justifications exist so far. (That is, a protocol may be secure in a symbolic model, and yet a feasible attack exists.) In this talk, we will study a recent, symbolic definition of security against guessing attacks, based on static equivalence. First, we will present a decision procedure for a large class of protocols, for a finite number of sessions. Then, we will provide a computational justification in the case of a passive adversary, that is, a pure eavesdropper. [This part is a joint work with M. Abadi and B. Warinschi.]