Evaluation of Claude Mythos Preview's cyber capabilities
aisi.gov.uk
April 13, 2026
1 min read
🔥🔥🔥🔥🔥
43/100
Summary
Claude Mythos Preview shows improved performance in capture-the-flag (CTF) challenges and significant advancements in multi-step cyber-attack simulations. The evaluation indicates enhanced cyber capabilities compared to previous versions.
Community Sentiment
Positives
- Mythos is the first model to complete all steps of 'The Last Ones' evaluation, showcasing significant advancements in automated network takeover capabilities.
- The evaluation indicates continued improvement in capture-the-flag challenges and notable enhancements in multi-step cyber-attack simulations, suggesting a positive trajectory for AI in cybersecurity.
Concerns
- Despite some improvements, the performance metrics suggest that Mythos only marginally outperforms previous models like Opus 4.6, raising questions about its overall impact.
- The lack of active defenders in the evaluation raises concerns about the real-world applicability of the results, as actual networks have security teams that could thwart such attacks.
- The absence of confidence intervals in the evaluation undermines the claims of significant improvement, highlighting a need for more rigorous statistical analysis in AI assessments.
![System Card: Claude Mythos Preview [pdf]](https://pub-90f0ac00d93c47daac3e1d2cdd28d496.r2.dev/articles/7b8a89e8454a0bcc169dd0040a75d6e6.webp)