Some thoughts about Anthropic's new cryptanalysis results
blog.cryptographyengineering.com
July 29, 2026
11 min read
🔥🔥🔥🔥🔥
49/100
Summary
Anthropic published two cryptanalysis results using Claude Mythos, targeting the HAWK signature scheme and presenting an improved attack on reduced-round AES. A blog post detailing the research process was also released alongside these findings.
Key Takeaways
- Anthropic published two new cryptanalysis results generated by their unreleased model, Claude Mythos, targeting the HAWK signature scheme and reduced-round AES.
- The attack on HAWK reduces its security by roughly half, but does not break the deployed version; it can be mitigated by increasing key sizes.
- The attack on reduced-round AES is less significant, as it targets a known weaker variant of the cipher that has been previously attacked.
- The HAWK attack demonstrates that existing cryptographic tools can be effectively applied, highlighting the potential of AI in cryptanalysis.
Community Sentiment
Positives
- The models are getting better at a fast clip, showing measurable and impressive progress in just five months — it's a clear sign that AI's capabilities are evolving rapidly.
- LLMs are like search engines on steroids, acting as a research librarian that opens up vast oceans of human knowledge, vastly expanding our intellectual reach.
Concerns
- There's a concerning trend of dismissing these models as 'glorified autocomplete' — it shows a lack of understanding of their true capabilities and the advancements being made.
- The filtering in Fable downgrades useful inquiries, indicating that useful knowledge might be unnecessarily restricted, which could stifle innovation in important fields.
