Anthropic's AI Exposes Critical Flaw in HAWK Post-Quantum Algorithm
Anthropic's AI model, Claude Mythos Preview, has revealed a significant weakness in HAWK, a post-quantum digital signature algorithm designed to secure online transactions. After analyzing HAWK for approximately 60 hours at a computing cost of around $100,000, Claude identified an attack that reduces the effective key strength from 2^64 operations to 2^38 operations.
The discovery highlights the potential of advanced AI in accelerating cryptographic research and shifting the focus towards quantum-resistant security. HAWK had previously passed two years of expert review and multiple rounds of human evaluation, but Claude's analysis revealed a vulnerability that could compromise its effectiveness.
Anthropic emphasized that their findings do not affect Bitcoin or Ethereum, as HAWK is not part of their current or proposed security framework. Instead, both cryptocurrencies rely on three quantum-safe algorithms standardized by the U.S. National Institute of Standards and Technology (NIST) for their proposed quantum-resistance roadmap.
The announcement comes at a time when blockchain protocols are increasingly adopting zero-knowledge proof systems, which utilize hash functions like Poseidon to ensure privacy and security. However, Claude's AI achieved only a less-than-tenfold improvement against Poseidon, indicating that more work is needed in this area.