Anthropic's Claude Mythos Uncovers Vulnerabilities in Post-Quantum HAWK and Reduced AES-128
Anthropic researchers used the closed Claude Mythos Preview model to find two cryptanalytic attacks on the HAWK digital signature scheme and a shortened AES-128 version. The findings have no practical impact but highlight expert-level AI research capabilities.
Anthropic's internal research model, Claude Mythos Preview, helped identify two weaknesses during cryptographic analysis. The first targets HAWK, a post-quantum digital signature scheme, while the second affects a reduced-round variant of the AES-128 cipher.
"These are significant research achievements but have no practical impact on today's systems," Anthropic stated. The company emphasized that the results do not compromise any production software and require no immediate patches or changes.
The discovery underscores how advanced AI models can now contribute to cryptography research at an expert level, though the findings remain confined to non-deployed or weakened versions of algorithms.
Source: ForkLog