Mythos AI finds critical flaw in quantum-safe algorithm HAWK; developer withdraws it
Anthropic's security model Mythos discovered an unknown vulnerability in the third-round PQC algorithm HAWK, rendering it broken. The algorithm's developer withdrew it from NIST consideration.
AI discovery cripples HAWK algorithm
Anthropic announced that its AI security model Mythos has found a serious weakness in HAWK, a digital signature scheme designed to withstand quantum computer attacks. HAWK was in NIST's third round of testing for post-quantum cryptography (PQC) when Mythos discovered a method that effectively halves its key strength. With about 60 hours of work and $100,000 in compute costs, a researcher with no cryptography expertise used Mythos to improve the best-known attack by exploiting automorphism symmetries—a previously unused approach.
The mathematical foundation of HAWK (the Lattice Isomorphism Problem) was bypassed by finding a new way to detect automorphism symmetries. While the vulnerability can be mitigated by doubling the key size, the added computation makes HAWK less attractive compared to existing PQC signature algorithms like ML-DSA and FN-DSA.
Johns Hopkins professor Matthew Green noted that the attack does not invent fundamentally new math but combines existing tools. Google PQC expert Sophie Schmieg concluded: “Basically with this paper, HAWK is dead.”
Improvements to AES attack
Mythos also enhanced a meet-in-the-middle attack on AES. Using a Möbius Bridge technique, it reduced the required plaintext inputs from 2^105 to 2^89, cutting attack time by a factor of 200 to 800. However, these results were obtained on a weakened AES version (7 rounds instead of 10–14) and are infeasible outside a lab.
Caveats and outlook
Anthropic acknowledges both findings are incremental and do not break any widely used cryptosystems. Mythos is currently available only to a limited set of users. It is unknown whether researchers using traditional methods were close to these discoveries. Still, AI's role in finding cryptographic weaknesses is growing and may significantly impact the security field in the future.
