Anthropic’s latest AI, Claude Mythos Preview, uncovered significant vulnerabilities in cryptographic systems believed to be quantum-secure. The AI's autonomous research exposed new weaknesses in HAWK, a post-quantum digital signature scheme, and a reduced-round variant of AES, shaking long-held confidence in these algorithms.
Claude's discovery in HAWK centers on a previously unseen mathematical symmetry, slashing the computational effort required to break its smallest key from 2^64 to 2^38 operations. This dramatic drop about 67 million times less work exposes a glaring flaw in the scheme that had withstood two years of expert human analysis. Fixing it would mean doubling the key size, a move that undermines the speed and efficiency benefits that originally made HAWK attractive for quantum resistance.
AI Disrupts Cryptography and Sparks Debate on Bitcoin’s Quantum Readiness
The AI also delivered a breakthrough against a trimmed-down, seven-round version of AES-128, accelerating attack speeds by up to 800 times. While the full AES-128 algorithm remains secure, Claude introduced a novel "Möbius Bridge" technique that eliminates a guessing step in the crypto analysis. Researchers confirmed these results only after nearly a month of validation following Claude’s three days of autonomous work.
Importantly, Anthropic assures these vulnerabilities pose no immediate threat to current internet security, as HAWK is not yet deployed and the AES exposure involves a non-standard version. However, the findings shows AI’s growing ability to detect hidden weaknesses in encryption, a trend that could reshape cybersecurity.
Samson Mow, CEO of JAN3, weighed in with a cautious perspective on Bitcoin’s adoption of post-quantum cryptography. He emphasized that such discoveries reinforce the need for deliberate, measured transitions rather than rushing into quantum-resistant signatures that might carry unforeseen risks. His stance aligns with the growing debate over balancing innovation with security in the blockchain space.
