Anthropic’s AI model Mythos found a fundamental weakness in HAWK, a post-quantum cryptography (PQC) candidate algorithm, Ars Technica reported on July 29 local time. After the finding was confirmed, developers voluntarily withdrew HAWK from the NIST standardisation process. Mythos improved on existing attack methods for HAWK and cut the effective key strength in half, so the algorithm no longer met the required security level. Mythos did not create a new mathematical method but combined existing cryptographic techniques in ways researchers had not tried. Matthew Green, a cryptography professor at Johns Hopkins University, assessed this as particularly noteworthy. Anthropic stressed the result does not mean current cryptographic standards have been broken, as the finding was verified using a specially prepared test version for research purposes, and the attack itself requires enormous computing power, making it impossible to carry out outside a laboratory. Even so, it said it views the result as a sign that large language models are rapidly advancing in codebreaking capability.