Anthropic AI breaks leading post-quantum crypto candidate
Anthropic's Claude Mythos AI model has exposed a fundamental vulnerability in the HAWK post-quantum encryption algorithm, casting doubt on the crypto industry's transition to quantum-resistant security.
Anthropic has revealed that an unreleased version of its Claude model, Claude Mythos Preview, successfully compromised HAWK. HAWK is a digital signature system competing to become a US federal standard that proves a transaction originated from a specific user without exposing their private key. The AI discovered a previously unknown mathematical symmetry that drastically reduced the computing power required to steal the algorithm's smallest key. Anthropic reported the attack dropped the cost from 2^64 operations to 2^38, making the key roughly 67 million times easier to breach.
The algorithm was specifically designed to survive attacks from future quantum computers. It had advanced to the third round of the National Institute of Standards and Technology's post-quantum competition in May, standing as the final lattice-based candidate in the running. However, repairing the vulnerability requires roughly doubling HAWK's key sizes. As Anthropic noted, this fix "eliminates many of the reasons" the algorithm was considered attractive, directly undermining its core pitch of compact keys and fast signing.
In blockchain networks, signature size equates to block space, and block space directly dictates transaction fees. Any cryptocurrency network shopping for a quantum-resistant replacement is essentially selecting based on bytes per signature. By inflating the key sizes, the necessary patch effectively neutralizes the efficiency edge that made HAWK a leading contender for widespread commercial adoption.
To be clear, HAWK has never been deployed on any live network. Bitcoin and other major cryptocurrencies still operate on ECDSA, the pre-quantum scheme that candidates like HAWK are eventually intended to replace. No existing blockchain or crypto holding is currently at risk from this specific flaw.
Nevertheless, the development introduces a significant risk factor for the digital asset industry. Post-quantum security remains the largest long-term threat hanging over the sector. The entire quantum defense infrastructure built by the industry over recent months depends on the assumption that next-generation cryptographic schemes are fundamentally secure.
Major financial institutions and blockchain developers have recently accelerated these efforts. A consortium backed by BlackRock, Coinbase, and Strategy has formed around quantum readiness. Parallel roadmaps have been published by Ethereum co-founder Vitalik Buterin, as well as by the teams behind the NEAR and Zcash networks.
Claude Mythos has now demonstrated that a leading candidate in this defensive migration contained a critical flaw before it ever shipped. Anthropic observed that the same AI tools capable of breaking encryption can also be used to audit and defend it, which is why firms are racing to point AI at their own codebases. For market participants, the immediate takeaway is whether artificial intelligence will outpace the industry's ability to rebuild secure cryptographic foundations.
