AI Discovers Cryptographic Weaknesses in HAWK and AES

Researchers at Anthropic utilized Claude Mythos Preview to uncover significant vulnerabilities in cryptographic algorithms, specifically targeting the HAWK digital signature scheme and a reduced-round version of the Advanced Encryption Standard (AES). The attack on HAWK, a candidate for post-quantum cryptography, effectively halves its key strength, necessitating larger key sizes to maintain security. The AES vulnerability, while theoretical, allows for a much faster attack on a simplified version of the cipher, but does not affect the full 10-round AES used in practice. Both findings do not currently compromise any deployed systems, but they highlight the potential for AI to discover flaws previously overlooked by human experts. The research was disclosed to relevant parties, including NIST, prior to public announcement. The implications of these findings could affect billions of users if similar flaws are found in widely used cryptographic systems.

Key Points: • Claude Mythos Preview found critical vulnerabilities in HAWK and AES. • HAWK's key strength is effectively halved, requiring larger keys for security. • The AES attack is theoretical and does not affect current production systems.