Evaluation of Anthropic's New AI Cryptanalysis Results

Anthropic's unreleased advanced model Claude Mythos has shown notable achievements in the field of cryptanalysis. This model derived improvements to the HAWK key recovery attack and the 7-round AES attack, demonstrating an ability that goes beyond simply understanding existing cryptanalysis techniques to combining and extending them.

The key findings are as follows:

• HAWK Attack: While not at a level that immediately breaks currently deployed cryptographic schemes, it showed an effect of reducing the number of security bits by roughly half. Accordingly, key sizes would need to be doubled to maintain the same level of security strength.
• AI's Capability: These results suggest that AI can combine complex mathematical/cryptographic logic to find and advance vulnerabilities in existing security schemes.