Anthropic's Claude Mythos Finds Weaknesses in Post-Quantum Cryptography
The research firm Anthropic has made significant advancements in post-quantum cryptography using its Claude Mythos AI model. According to the company, the AI model discovered two potential vulnerabilities in existing cryptographic algorithms: a key recovery attack on the HAWK digital signature scheme and an attack on a reduced-round AES-128 variant.
The researchers used the Claude Mythos Preview model to identify the weaknesses, which were then verified through extensive research. Anthropic emphasized that these findings do not affect current production systems and do not require changes to deployed software.
The company's results demonstrate the potential of AI in cryptography analysis, allowing for more efficient testing of cryptographic algorithms before they are deployed. However, as model capabilities grow, new approaches to responsible disclosure of potentially critical vulnerabilities will be needed.