Pierluigi Paganini
July 29, 2026
Anthropic published two cryptographic research results achieved by Claude Mythos Preview working mostly autonomously: an improved attack on HAWK, a post-quantum digital signature scheme currently under NIST review, and a 200 to 800 times faster attack on a reduced-round version of AES. Neither result breaks any production system today. Both results show that frontier AI can now do original cryptographic research at a level that two years of human expert review failed to achieve.
“When we launched Claude Mythos Preview, we showed it was able to autonomously find and exploit vulnerabilities in almost every piece of software we pointed it at. This included several major cryptographic libraries—shared collections of code that are used to encrypt data.” reads Anthropic’s announcement.
“The vulnerabilities that Claude found in these cryptographic libraries1 were due to incorrect implementation of the algorithms—that is, errors in how programmers used the algorithms in their code that created opportunities for attackers to break the encryption. Now, we have found that Claude is able to find mathematical flaws in the algorithms themselves.”
The HAWK result is the more immediately significant of the two. HAWK is one of the remaining third-round candidates in NIST’s ongoing post-quantum cryptography standardization process, which has been running for nearly a decade. Mythos found a previously unexploited symmetry in the mathematical structure underlying HAWK, a nontrivial automorphism, that allows a faster enumeration attack and effectively cuts the scheme’s key strength in half. For the smallest HAWK-256 key size, the expected attack cost drops from 2^64 operations to 2^38. Doubling HAWK’s key size would restore the original security level, but that eliminates most of the practical advantages that made the scheme an attractive post-quantum candidate in the first place.
The HAWK result took 60 hours, coordinated by one Anthropic researcher who was not an expert in lattice-based cryptography.
“To find the attack, Claude Mythos Preview worked semi-autonomously in an agentic harness, with occasional human guidance and nontechnical direction. Mythos found the attack after an extensive literature review to understand the state of the art, and substantial mathematical reasoning and computational experiments.” continues the announcement. “After finding the attack, Mythos implemented an end-to-end verification pipeline to convince itself—and the human operator—of the attack’s correctness.”
The multi-agent workflow produced an interesting dynamic: two worker agents investigated the same idea in parallel. The first rejected it as infeasible. The second found a way to exploit it fully. They kept exchanging messages until both agreed they had found something real. Total API cost: approximately $100,000.
The AES result took longer and cost far more human time to validate than to discover. Mythos worked for three days generating several hundred million output tokens and produced what Anthropic calls a Möbius Bridge, a fingerprinting technique that eliminates one of the guesses an attacker needs to make in a meet-in-the-middle attack on 7-round AES, then compensates for the added computational cost with several optimization techniques. The attack doesn’t touch the full 10-round AES-128 used in production. The discovery was almost entirely autonomous; the researcher’s interventions amounted to three substantive prompts over three days, one of which simply said the model needed to keep searching for genuinely novel ideas rather than settling for incremental gains.
“Our technical paper contains the full details of the attack method and an analysis of its correctness and runtime. Compared to the one week that Mythos spent conceiving the idea, the vast majority of human researchers’ time was spent validating the correctness of its claims (though it is important to note the researchers are not experts in cryptography).” Anthropic states.
Two researchers spent close to a month gaining enough confidence in the result to write a paper about it.
Beyond the two main results, Mythos also developed a practical attack recovering a 13-round LEA key in under 2^30 encrypted plaintexts running in under an hour on a modern desktop, dramatically better than the prior best of 2^98 plaintext pairs and 2^86 work. Preliminary results also cover 6-round Serpent-128, with limited improvements on Salsa20, the Poseidon hash function, and SHA-1. Anthropic followed responsible disclosure throughout, sharing the HAWK finding with the scheme’s authors in June before coordinating public disclosure to the NIST mailing list alongside the paper’s release.
The broader implication is the one Anthropic states plainly: in one year, language models went from being unable to perform cryptanalysis of basic ciphers to finding flaws in designs that escaped years of expert human review. The cybersecurity community is already struggling to keep up with the volume of bugs AI can find. Anthropic predicts the same bottleneck is coming for academic cryptography research, where human researchers may increasingly find themselves spending months validating results that a model produced in days.
“Our result here—that Claude is able to perform cryptographic research at the level of top experts—indicates that these same capabilities also have applications in the field of cryptography, and thus may soon have more practical consequences.” concludes the company. “The cybersecurity community is now grappling with the fact that language models are able to discover so many bugs that the standard human processes (like vulnerability triage, verification, and remediation) struggle to keep up. We predict that the same will soon be true in academic cryptography research. “
Follow me on Twitter: @securityaffairs and Facebook and Mastodon
(SecurityAffairs – hacking, Anthropic)
