Anthropic has announced a significant breakthrough in cryptographic security, revealing that its AI model Claude Mythos Preview identified previously undetected mathematical flaws in two widely-used encryption algorithms. These findings highlight the potential of AI to uncover vulnerabilities that human experts may have overlooked for years.
HAWK Algorithm Severely Weakened
The first discovery involves HAWK, a post-quantum digital signature scheme currently under review by the National Institute of Standards and Technology (NIST). According to Anthropic, Claude Mythos Preview demonstrated that HAWK’s effective key strength is significantly lower than previously believed, essentially halving its security margin. This is particularly concerning as HAWK is being considered for standardization in a future where quantum computers could break traditional encryption methods.
Improved Attack on AES
The second finding relates to AES, the most commonly used symmetric encryption algorithm globally. Claude Mythos Preview improved the best-known attack on a seven-round version of AES, achieving a 200- to 800-fold increase in efficiency. While the full AES-256 standard remains secure, this advancement could have implications for reduced-round versions used in specific applications, raising questions about the robustness of certain implementations.
Implications for Cybersecurity
These discoveries underscore the growing role of AI in cybersecurity. As encryption systems become more complex, traditional analysis methods may fall short. Anthropic's findings suggest that AI tools like Claude could serve as powerful allies in identifying and mitigating security risks before they are exploited. The revelations also emphasize the need for continuous scrutiny of cryptographic standards, especially as quantum computing advances.
While the impact on full implementations of these algorithms remains limited, the findings are a wake-up call for the cybersecurity community to re-evaluate existing protocols and ensure robust defenses against evolving threats.
