Security and Privacy in RFID Systems

This PhD thesis is concerned with authentication protocols using portable lightweight devices such as RFID tags. these devices have lately gained a significant attention for the diversity of the applications that could benefit form their features, ranging from inventory systems and building access control, to medical devices. However, the emergence of this technology has raised concerns about the possible loss of privacy carrying such tags induce in allowing tracing persons or unveiling the contents of a hidden package. this fear led to the appearance of several organizations which goal is to stop the spread of RFID tags. We take a cryptographic viewpoint on the issue and study the extent of security and privacy that RFID-based solutions can offer. In the first part of this thesis, we concentrate on analyzing two original primitives that were proposed to ensure security for RFID tags. the first one, HB#, is a dedicated authentication protocol that exclusively uses very simple arithmetic operations: bitwise AND and XOR. HB# was proven to be secure against a certain class of man-in-the-middle attacks and conjectured secure against more general ones. We show that the latter conjecture does not hold by describing a practical attack that allows an attacker to recover the tag's secret key. Moreover, we show that to be immune against our attack, HB#'s secret key size has to be increased to be more than 15 000 bits. this is an unpractical value for the considered applications. We then turn to SQUASH, a message authentication code built around a public-key encryption scheme, namely Rabin's scheme. By mounting a practical key recovery attack on the earlier version of SQUASH, we show that the security of all versions of SQUASH is unrelated to the security of Rabin encryption function. The second part of the thesis is dedicated to the privacy aspects related to the RFID technology. We first emphasize the importance of establishing a framework that correctly captures the intuition that a privacy-preserving protocol does not leak any information about its participants. For that, we show how several protocols that were supported by simple arguments, in contrast to a formal analysis, fail to ensure privacy. Namely, we target ProbIP, MARP, Auth2, YA-TRAP, YA-TRAP+, O-TRAP, RIPP-FS, and the Lim-Kwon protocol. We also illustrate the shortcomings of other privacy models such as the LBdM model. The rest of the dissertation is then dedicated to our privacy model. Contrarily to most RFID privacy models that limit privacy protection to the inability of linking the identity of two participants in two different protocol instances, we introduce a privacy model for RFID tags that proves to be the exact formalization of the intuition that a private protocol should not leak any information to the adversary. the model we introduce is a refinement of Vaudenay's one that invalidates a number of its limitations. Within these settings, we are able to show that the strongest notion of privacy, namely privacy against adversaries that have a prior knowledge of all the tags' secrets, is realizable. To instantiate an authentication protocol that achieves this level of privacy, we use plaintext-aware encryption schemes. We then extend our model to the case of mutual authentication where, in addition to a tag authenticating to the reader, the reverse operation is also required.