Provably Secure Authenticated Encryption

Authenticated Encryption (AE) is a symmetric key cryptographic primitive that ensures confidentiality and authenticity of processed messages at the same time. The research of AE as a primitive in its own right started in 2000.

The security goals of AE were captured in formal definitions in the tradition in the tradition of provable security (such as NAE, MRAE, OAE, RAE or the RUP), where the security of a scheme is formally proven assuming the security of an underlying building block. The prevailing syntax moved to nonce-based AE with associated data (which is an additional input that gets authenticated, but not encrypted). Other types of AE schemes appeared as well, e.g. ones that supported stateful sessions. Numerous AE schemes were designed; in the early years, these were almost exclusively blockcipher modes of operation, most notably OCB in 2001, CCM in 2003 and GCM in 2004. At the same time, issues were discovered both with the security and applicability of the most popular AE schemes, and other applications of symmetric key cryptography.

As a response, the Competition for Authenticated Encryption: Security, Applicability, and Robustness (CAESAR) was started in 2013. Its goals were to identify a portfolio of new, secure and reliable AE schemes that would satisfy the needs of practical applications, and also to boost the research in the area of AE. Prompted by CAESAR, 57 new schemes were designed, new types of constructions that gained popularity appeared (such as the Sponge-based AE schemes), and new notions of security were proposed (such as RAE). The final portfolio of the CAESAR competition should be announced in 2018.

In this thesis, we push the state of the art in the field of AE in several directions. All of them are related to provable security, in one way, or another. We propose OMD, the first provably secure dedicated AE scheme that is based on a compression function. We further modify OMD to achieve nonce misuse-resistant security (MRAE). We also propose another provably secure variant of OMD called pure OMD, which enjoys a great improvement of performance over OMD. Inspired by the modifications that gave rise to pure OMD, we turn to the popular Sponge-based AE schemes and prove that similar measures can also be applied to the keyed Sponge and keyed Duplex (a variant of the Sponge), allowing a substantial increase of performance without an impact on security.

We then address definitional aspects of AE. We critically evaluate the security notion of OAE, whose authors claimed that it provides the best possible security for online schemes under nonce reuse. We challenge these claims, and discuss what are the meaningful requirements for online AE schemes. Based on our findings, we formulate a new definition of online AE security under nonce-reuse, and demonstrate its feasibility. We next turn our attention to the security of nonce-based AE schemes under stretch misuse; i.e. when a scheme is used with varying ciphertext expansion under the same key, even though it should not be. We argue that varying the stretch is plausible, and formulate several notions that capture security in presence of variable stretch. We establish their relations to previous notions, and demonstrate the feasibility of security in this setting.

We finally depart from provable security, with the intention to complement it. We compose a survey of universal forgeries, decryption attacks and key recovery attacks on 3rd round CAESAR candidates.

Secure communication using authenticated channels

Sylvain Pasini

Our main motivation is to design more user-friendly security protocols. Indeed, if the use of the protocol is tedious, most users will not behave correctly and, consequently, security issues occur. An example is the actual behavior of a user in front of an SSH certificate validation: while this task is of utmost importance, about 99% of SSH users accept the received certificate without checking it. Designing more user-friendly protocols may be difficult since the security should not decrease at the same time. Interestingly, insecure channels coexist with channels ensuring authentication. In practice, these latters may be used for a string comparison or a string copy, e.g., by voice over IP spelling. The shorter the authenticated string is, the less human interaction the protocol requires, and the more user-friendly the protocol is. This leads to the notion of SAS-based cryptography, where SAS stands for Short Authenticated String. In the first part of this thesis, we analyze and propose optimal SAS-based message authentication protocols. By using these protocols, we show how to construct optimal SAS-based authenticated key agreements. Such a protocol enables any group of users to agree on a shared secret key. SAS-based cryptography requires no pre-shared key, no trusted third party, and no public-key infrastructure. However, it requires the user to exchange a short SAS, e.g., five decimal digits. By using the just agreed secret key, the group can now achieve a secure communication based on symmetric cryptography. SAS-based authentication protocols are often used to authenticate the protocol messages of a key agreement. Hence, each new secure communication requires the interaction of the users to agree on the SAS. A solution to reduce the user interaction is to use digital signature schemes. Indeed, in a setup phase, the users can use a SAS-based authentication protocol to exchange long-term verification keys. Then, using digital signatures, users are able to run several key agreements and the authentication of protocol messages is done by digital signatures. In the case where no authenticated channel is available, but a public-key infrastructure is in place, the SAS-based setup phase is avoided since verification keys are already authenticated by the infrastructure. In the second part of this thesis, we also study two problems related to digital signatures: (1) the insecurity of digital signature schemes which use weak hash functions and (2) the privacy issues from signed documents. Digital signatures are often proven to be secure in the random oracle model. The role of random oracles is to model ideal hash functions. However, real hash functions deviate more and more from this idealization. Indeed, weaknesses on hash functions have already been discovered and we are expecting new ones. A question is how to fix the existing signature constructions based on these weak hash functions. In this thesis, we first try to find a better way to model weak hash function. Then, we propose a (randomized) pre-processing to the input message which transforms any weak signature implementation into a strong signature scheme. There remains one drawback due to the randomization. Indeed, the random coins must be sent and thus the signature enlarges. We also propose a method to avoid the increase in signature length by reusing signing coins. Digital signatures may also lead to privacy issues. Indeed, given a message and its signature, anyone can publish the pair which will confirm the authenticity of the message. In certain applications, like in electronic passports (e-passports), publishing the authenticated data leads to serious privacy issues. In this thesis, we define the required security properties in order to protect the data privacy, especially in the case of e-passport verification. The main idea consists for the e-passport to keep the signature secret. The e-passport should only prove that it knows a valid signature instead of revealing it. We propose a new primitive, called Offline Non-Transferable Authentication Protocol (ONTAP), as well as efficient implementations that are compatible with the e-passport standard signature schemes.

EPFL2009

Bringing Theory Closer to Practice in Post-quantum and Leakage-resilient Cryptography

Alexandre Raphaël Duc

Modern cryptography pushed forward the need of having provable security. Whereas ancient cryptography was only relying on heuristic assumptions and the secrecy of the designs, nowadays researchers try to make the security of schemes to rely on mathematical problems which are believed hard to solve. When doing these proofs, the capabilities of potential adversaries are modeled formally. For instance, the black-box model assumes that an adversary does not learn anything from the inner-state of a construction. While this assumption makes sense in some practical scenarios, it was shown that one can sometimes learn some information by other means, e.g., by timing how long the computation take. In this thesis, we focus on two different areas of cryptography. In both parts, we take first a theoretical point of view to obtain a result. We try then to adapt our results so that they are easily usable for implementers and for researchers working in practical cryptography. In the first part of this thesis, we take a look at post-quantum cryptography, i.e., at cryptographic primitives that are believed secure even in the case (reasonably big) quantum computers are built. We introduce HELEN, a new public-key cryptosystem based on the hardness of the learning from parity with noise problem (LPN). To make our results more concrete, we suggest some practical instances which make the system easily implementable. As stated above, the design of cryptographic primitives usually relies on some well-studied hard problems. However, to suggest concrete parameters for these primitives, one needs to know the precise complexity of algorithms solving the underlying hard problem. In this thesis, we focus on two recent hard-problems that became very popular in post-quantum cryptography: the learning with error (LWE) and the learning with rounding problem (LWR). We introduce a new algorithm that solves both problems and provide a careful complexity analysis so that these problems can be used to construct practical cryptographic primitives. In the second part, we look at leakage-resilient cryptography which studies adversaries able to get some side-channel information from a cryptographic primitive. In the past, two main disjoint models were considered. The first one, the threshold probing model, assumes that the adversary can put a limited number of probes in a circuit. He then learns all the values going through these probes. This model was used mostly by theoreticians as it allows very elegant and convenient proofs. The second model, the noisy-leakage model, assumes that every component of the circuit leaks but that the observed signal is noisy. Typically, some Gaussian noise is added to it. According to experiments, this model depicts closely the real behaviour of circuits. Hence, this model is cherished by the practical cryptographic community. In this thesis, we show that making a proof in the first model implies a proof in the second model which unifies the two models and reconciles both communities. We then look at this result with a more practical point-of-view. We show how it can help in the process of evaluating the security of a chip based solely on the more standard mutual information metric.

EPFL2015

Cryptographic Hash Functions in Groups and Provable Properties

Juraj Sarinay

We consider several "provably secure" hash functions that compute simple sums in a well chosen group (G,*). Security properties of such functions provably translate in a natural way to computational problems in G that are simple to define and possibly also hard to solve. Given k disjoint lists Li of group elements, the k-sum problem asks for gi ∊ Li such that g1 * g2 ... gk = 1G. Hardness of the problem in the respective groups follows from some "standard" assumptions used in public-key cryptology such as hardness of integer factoring, discrete logarithms, lattice reduction and syndrome decoding. We point out evidence that the k-sum problem may even be harder than the above problems. Two hash functions based on the group k-sum problem, SWIFFTX and FSB, were submitted to NIST as candidates for the future SHA-3 standard. Both submissions were supported by some sort of a security proof. We show that the assessment of security levels provided in the proposals is not related to the proofs included. The main claims on security are supported exclusively by considerations about available attacks. By introducing "second-order" bounds on bounds on security, we expose the limits of such an approach to provable security. A problem with the way security is quantified does not necessarily mean a problem with security itself. Although FSB does have a history of failures, recent versions of the two above functions have resisted cryptanalytic efforts well. This evidence, as well as the several connections to more standard problems, suggests that the k-sum problem in some groups may be considered hard on its own, and possibly lead to provable bounds on security. Complexity of the non-trivial tree algorithm is becoming a standard tool for measuring the associated hardness. We propose modifications to the multiplicative Very Smooth Hash and derive security from multiplicative k-sums in contrast to the original reductions that related to factoring or discrete logarithms. Although the original reductions remain valid, we measure security in a new, more aggressive way. This allows us to relax the parameters and hash faster. We obtain a function that is only three times slower compared to SHA-256 and is estimated to offer at least equivalent collision resistance. The speed can be doubled by the use of a special modulus, such a modified function is supported exclusively by the hardness of multiplicative k-sums modulo a power of two. Our efforts culminate in a new multiplicative k-sum function in finite fields that further generalizes the design of Very Smooth Hash. In contrast to the previous variants, the memory requirements of the new function are negligible. The fastest instance of the function expected to offer 128-bit collision resistance runs at 24 cycles per byte on an Intel Core i7 processor and approaches the 17.4 figure of SHA-256. The new functions proposed in this thesis do not provably achieve a usual security property such as preimage or collision resistance from a well-established assumption. They do however enjoy unconditional provable separation of inputs that collide. Changes in input that are small with respect to a well defined measure never lead to identical output in the compression function.

EPFL2011