ANTHROPIC'S Mythos AI detected vulnerabilities in cryptographic algorithms, revealing weaknesses in two systems: HAWK and AES. The findings suggest methods to reduce the work needed to compromise these systems but do not break existing cryptographic standards. For HAWK, Mythos halved the key strength through an innovative attack after approximately 60 hours of computation, raising concerns about its viability against future quantum threats.
Meanwhile, a less dramatic improvement for AES involved a refined meet-in-the-middle attack technique that greatly decreased the required inputs for a successful breach. Despite the advances, both attacks are considered impractical outside controlled environments. Critically, the report highlights the potential of AI in cryptanalysis while cautioning against overestimating its immediate applicability.