HAWK post-quantum algorithm withdrawn after Mythos attack halves key

The HAWK digital signature scheme was withdrawn from the NIST post-quantum cryptography standardization process on July 28, 2026, a day after Anthropic revealed that its Mythos AI model had halved the algorithm’s effective key strength.

HAWK had been undergoing third-round evaluation for algorithms designed to withstand quantum-computer attacks. Post-quantum cryptography aims to secure communications against machines far more powerful than today’s computers.

Anthropic said a researcher without cryptography expertise used Mythos semi-autonomously for about 60 hours at a cost of roughly $100,000 to improve the best-known attack. The model worked through literature review, mathematical reasoning and computational experiments, deploying two independent agents that eventually collaborated.

The attack targets the Lattice Isomorphism Problem by finding automorphism symmetries more efficiently. Sophie Schmieg, a post-quantum cryptography expert at Google, said the finding makes HAWK less competitive than alternatives like ML-DSA and FN-DSA. “With this paper, HAWK is dead,” she wrote.

Mythos also achieved a modest improvement against a weakened, 7-round version of the AES cipher. The technique reduced the required plaintext inputs from 2^105 to 2^89, a reduction that could speed key recovery 200- to 800-fold but remains infeasible in practice.

Johns Hopkins professor Matthew Green noted that the HAWK attack combined existing tools in a novel way, a task that suits AI. Anthropic stressed that both attacks used challenge instances and are not feasible in real-world settings.

The company plans to continue refining Mythos, currently available only to a select group. It predicts language models will increasingly produce novel cryptanalytic results, potentially overwhelming human researchers.