International Association
for Cryptologic Research

Given the recent progress in machine learning (ML), the cryptography community has started exploring the applicability of ML methods to the design of new cryptanalytic approaches. While current empirical results show promise, the extent to which such methods may outperform classical cryptanalytic approaches is still somewhat unclear. In this work, we initiate exploration of the theory of ML-based cryptanalytic techniques, in particular providing new results towards understanding whether they are fundamentally limited compared to traditional approaches. Whereas most classic cryptanalysis crucially relies on directly processing individual samples (e.g., plaintext-ciphertext pairs), modern ML methods thus far only interact with samples via gradient-based computations that average a loss function over all samples. It is, therefore, conceivable that such gradient-based methods are inherently weaker than classical approaches. We introduce a unifying framework for capturing both ``sample-based'' adversaries that are provided with direct access to individual samples and ``gradient-based'' ones that are restricted to issuing gradient-based queries that are averaged over all given samples via a loss function. Within our framework, we establish a general feasibility result showing that any sample-based adversary can be simulated by a seemingly-weaker gradient-based one. Moreover, the simulation exhibits a nearly optimal overhead in terms of the gradient-based simulator's running time. Finally, we extend and refine our simulation technique to construct a gradient-based simulator that is fully parallelizable (crucial for avoiding an undesirable overhead for parallelizable cryptanalytic tasks), which is then used to construct a gradient-based simulator that executes the particular and highly useful gradient-descent method. Taken together, although the extent to which ML methods may outperform classical cryptanalytic approaches is still somewhat unclear, our results indicate that such gradient-based methods are not inherently limited by their seemingly restricted access to the provided samples.

A line of recent work has highlighted the importance of context commitment security, which asks that authenticated encryption with associated data (AEAD) schemes will not decrypt the same adversarially-chosen ciphertext under two different, adversarially-chosen contexts (secret key, associated data, and nonce). Despite a spate of recent attacks, many open questions remain around context commitment; most obviously nothing is known about the commitment security of important schemes such as CCM, EAX, and SIV. We resolve these open questions, and more. Our approach is to, first, introduce a new framework that helps us more granularly define context commitment security in terms of what portions of a context are adversarially controlled. We go on to formulate a new security notion, called context discoverability, which can be viewed as analogous to preimage resistance from the hashing literature. We show that unrestricted context commitment security (the adversary controls all of the two contexts) implies context discoverability security for a class of schemes encompassing most schemes used in practice. Then, we show new context discovery attacks against a wide set of AEAD schemes, including CCM, EAX, SIV, GCM, and OCB3, and, by our general result, this gives new unrestricted context commitment attacks against them. Finally, we explore the case of restricted context commitment security for the original SIV mode, for which no prior attack techniques work (including our context discovery based ones). We are nevertheless able to give a novel $\bigO(2^{n/3})$ attack using Wagner's k-tree algorithm for the generalized birthday problem.

We build the first construction of a partially oblivious pseudorandom function (POPRF) that does not rely on bilinear pairings. Our construction can be viewed as combining elements of the 2HashDH OPRF of Jarecki, Kiayias, and Krawczyk with the Dodis-Yampolskiy PRF. We analyze our POPRF’s security in the random oracle model via reduction to a new one-more gap strong Diffie-Hellman inversion assumption. The most significant technical challenge is establishing confidence in the new assumption, which requires new proof techniques that enable us to show that its hardness is implied by the q-DL assumption in the algebraic group model. Our new construction is as fast as the current, standards-track OPRF 2HashDH protocol, yet provides a new degree of flexibility useful in a variety of applications. We show how POPRFs can be used to prevent token hoarding attacks against Privacy Pass, reduce key management complexity in the OPAQUE password authenticated key exchange protocol, and ensure stronger security for password breach alerting services.

Authenticated encryption with associated data (AEAD) forms the core of much of symmetric cryptography, yet the standard techniques for modeling AEAD assume recipients have no ambiguity about what secret key to use for decryption. This is divorced from what occurs in practice, such as in key management services, where a message recipient can store numerous keys and must identify the correct key before decrypting. Ad hoc solutions for identifying the intended key are deployed in practice, but these techniques can be inefficient and, in some cases, have even led to practical attacks. Notably, to date there has been no formal investigation of their security properties or efficacy. We fill this gap by providing the first formalization of nonce-based AEAD that supports key identification (AEAD-KI). Decryption now takes in a vector of secret keys and a ciphertext and must both identify the correct secret key and decrypt the ciphertext. We provide new formal security definitions, including new key robustness definitions and indistinguishability security notions. Finally, we show several different approaches for AEAD-KI and prove their security.

Content moderation is crucial for stopping abusive and harassing messages in online platforms. Existing moderation mechanisms, such as message franking, require platform providers to be able to associate user identifiers to encrypted messages. These mechanisms fail in metadata-private messaging systems, such as Signal, where users can hide their identities from platform providers. The key technical challenge preventing moderation is achieving cryptographic accountability while preserving deniability.In this work, we resolve this tension with a new cryptographic primitive: asymmetric message franking (AMF) schemes. We define strong security notions for AMF schemes, including the first formal treatment of deniability in moderation settings. We then construct, analyze, and implement an AMF scheme that is fast enough to use for content moderation of metadata-private messaging.

Message franking enables cryptographically verifiable reporting of abusive messages in end-to-end encrypted messaging. Grubbs, Lu, and Ristenpart recently formalized the needed underlying primitive, what they call compactly committing authenticated encryption (AE), and analyze security of a number of approaches. But all known secure schemes are still slow compared to the fastest standard AE schemes. For this reason Facebook Messenger uses AES-GCM for franking of attachments such as images or videos.We show how to break Facebook’s attachment franking scheme: a malicious user can send an objectionable image to a recipient but that recipient cannot report it as abuse. The core problem stems from use of fast but non-committing AE, and so we build the fastest compactly committing AE schemes to date. To do so we introduce a new primitive, called encryptment, which captures the essential properties needed. We prove that, unfortunately, schemes with performance profile similar to AES-GCM won’t work. Instead, we show how to efficiently transform Merkle-Damgärd-style hash functions into secure encryptments, and how to efficiently build compactly committing AE from encryptment. Ultimately our main construction allows franking using just a single computation of SHA-256 or SHA-3. Encryptment proves useful for a variety of other applications, such as remotely keyed AE and concealments, and our results imply the first single-pass schemes in these settings as well.