Anthropic AI breaks HAWK-256 in 60 hours, raising Bitcoin concerns

An unreleased Anthropic AI model has broken a post-quantum signature scheme in about 60 hours, reopening a question Bitcoin developers have yet to answer: when should the network begin migrating to quantum-resistant cryptography?

The finding poses no immediate threat to Bitcoin. HAWK-256 was never used by the network, and no existing wallets are affected. But it changes an important assumption behind Bitcoin’s post-quantum roadmap: AI is making it faster and cheaper to test, and potentially break, advanced cryptographic systems.

What Mythos did to HAWK, and what it cost

On July 28, Anthropic announced that its Claude Mythos Preview model, which is being made available to a select group of approved users, discovered an attack on HAWK-256, the last lattice-based signature candidate still in NIST’s third round of post-quantum cryptography evaluation, which has not previously been documented.

The AI revealed a concealed symmetry in the design of HAWK, which experts overlooked, causing the algorithm’s effectiveness in terms of safety to decline by fifty percent as well as bringing down the anticipated amount of work needed from around 264 to 238 operations. Based on the research blog by Anthropic, the attack took approximately 60 hours and required about 100,000 dollars worth of computation, and it was carried out by a researcher without formal training in cryptography.

… Continue reading the full article at the original source below.