Beyond bug detection to cryptographic analysis... 'Mythos' found algorithm vulnerabilities

TL;DR AI
2 min readKey summary
Anthropic says Claude Sonnet Preview found new cryptographic weaknesses by analyzing the math behind the HAWK algorithm, not just implementation bugs.
The model also developed a faster attack method called “Möbius Bridge” against 7-round AES.
Researchers validated the results with human review and published the paper and benchmarks, while noting there is no immediate impact on deployed AES or real-world services.
The findings suggest AI may help discover deeper flaws in cipher design and reshape future cryptographic analysis and defense.
