Anthropic researchers using Claude Mythos Preview have identified novel attack methods against cryptographic algorithms, including a significant weakening of HAWK, a post-quantum digital signature scheme, and a new approach to attacking round-reduced AES, the most widely used symmetric cipher, the company announced on July 28, 2026 (anthropic.com). These advances represent important progress in cryptanalysis using AI but do not currently impact any production systems.
The discoveries emerged after Anthropic launched Claude Mythos Preview, an AI model capable of autonomously finding and exploiting vulnerabilities in software. The team applied Claude to major cryptographic libraries, focusing on both implementation errors and theoretical weaknesses. The first attack undermines HAWK, designed for security against quantum computers, while the second reveals a new vector against round-reduced versions of AES, a cornerstone of data encryption worldwide (anthropic.com).
These findings highlight the growing role of AI in cryptanalysis, demonstrating that advanced language models can accelerate the discovery of cryptographic weaknesses. HAWK and AES are critical to securing online communications and data privacy, making these insights relevant to the future of cybersecurity. While no current systems are affected, the research underscores the need for ongoing evaluation of cryptographic standards in the AI era (anthropic.com).
Anthropic’s report on July 28 details these cryptographic advances using Claude Mythos Preview, marking a notable step in AI-assisted security research and prompting further scrutiny of post-quantum and symmetric encryption methods.