AI Model Weakens Post-Quantum Cryptography Candidate HAWK in 60 Hours
The successful attack by Claude Mythos Preview highlights a growing AI threat to the algorithms intended to protect blockchains from quantum computers.
Anthropic's Claude Mythos Preview model has successfully identified an attack that reduced the effective key strength of HAWK, a proposed post-quantum digital-signature scheme. The breakthrough demonstrates that classical AI can uncover vulnerabilities in next-generation cryptographic candidates far faster than previously anticipated.
The attack was executed in approximately 60 hours, with compute and API costs totaling roughly $100,000. Following these findings, the HAWK algorithm was withdrawn from the NIST standardization process. Beyond HAWK, the model also discovered attacks on reduced-round AES, improving attack speeds by 200 to 800 times.
The Race for Quantum Resistance
Bitcoin developers are currently evaluating various post-quantum cryptographic (PQC) algorithms to protect the network from future quantum computers. These machines could theoretically break the Elliptic Curve Digital Signature Algorithm (ECDSA) currently used to secure Bitcoin addresses. While HAWK is not implemented in Bitcoin or Ethereum, it represented the type of candidate the industry must vet to ensure long-term security.
The Emerging AI Threat
This event suggests that the 'quantum threat' may be preceded or augmented by an 'AI threat.' The ability of a classical AI model to weaken a PQC candidate in a matter of days indicates that the window for a safe migration to quantum-resistant standards is narrower than expected. If the very algorithms designed to save the network from quantum computers are themselves vulnerable to AI-driven analysis, the industry faces a double-edged sword.
Implications for Bitcoin
For the Bitcoin network, this acceleration may necessitate more aggressive migration strategies. The current plan to transition to quantum-resistant signatures assumes that some candidate algorithms will inevitably break; however, the speed of the HAWK collapse suggests that the vetting process must be more rigorous. Developers must now account for AI's ability to automate the discovery of cryptographic flaws, potentially shortening the timeline for implementing new security standards.
What's Next
Industry observers are now watching how NIST and blockchain developers adjust their evaluation frameworks to include AI-driven stress testing. While the HAWK failure does not immediately jeopardize current assets, it serves as a critical warning that the security of future cryptographic standards cannot rely solely on traditional mathematical proofs, but must also withstand the capabilities of advanced large language models.