Mythos attack on 3rd-round PQC algorithm candidate puts it out of commission

by | Aug 19, 2026 | Technology

Mythos attack on 3rd-round PQC algorithm candidate puts it out of commission

A post-quantum cryptographic algorithm named HAWK has been removed from consideration as an official US standard following the discovery of a significant weakness by Anthropic’s Mythos AI security model. HAWK is a digital signature scheme designed to resist future attacks from quantum computers and had successfully passed two rounds of testing by the National Institute of Standards and Technology (NIST) before being evaluated in a third round specifically intended to identify such vulnerabilities. After Anthropic announced its findings earlier in the week, HAWK’s developer announced the withdrawal of the algorithm from the standardization process.

The Mythos model identified mathematical weaknesses in HAWK’s underlying security foundation, which relies on the Lattice Isomorphism Problem. Through approximately 60 hours of computational work costing around $100,000, the AI system discovered a previously unknown method for finding automorphism symmetries that effectively halved the algorithm’s key strength. This discovery resulted from combining several existing mathematical techniques in a novel way rather than developing fundamentally new mathematics. According to experts, while the weakness could be addressed by doubling the key size, the added computational burden makes HAWK less competitive than other available post-quantum cryptographic alternatives such as ML-DSA and FN-DSA.

Anthropologic separately tested Mythos against AES, a widely-used cipher, with less dramatic results. The model improved upon existing “meet-in-the-middle” attacks by reducing the number of required plaintext inputs from approximately 2^105 to 2^89, potentially accelerating such attacks by 200- to 800-fold. However, security experts emphasize important limitations: the tested versions were weakened variants of the official specifications, the underlying mathematical foundations remain secure, and the attacks would likely be impractical outside controlled testing environments. The findings highlight potential advances in cryptanalysis while underscoring that current cryptographic systems remain safe for present-day use.

Article Attribution | Read More at Article Source

Article summary produced by Claude AI