Detailed cyber evaluations of Claude 4

Brief overview

Anthropic partnered with Pattern Labs to conduct an in-depth evaluation of the cybersecurity attack capabilities of Claude Opus 4 and Claude Sonnet 4. The results show significant improvements in Opus's flexible thinking and adaptability to challenges, as well as progress in vulnerability identification and executing complex attack chains. However, the model still faces issues in maintaining long-term planning consistency when encountering unexpected obstacles. Pattern Labs has released the complete evaluation report.

Article

Anthropic (with Pattern Labs)

We believe we are at a crucial period for cybersecurity and AI, with models advancing toward human-level cyber offense capabilities in some scenarios. As part of our commitment to safety at the frontier, we conduct rigorous testing of our models' cyber offense capabilities. For Claude Opus 4 and Claude Sonnet 4, we partnered with Pattern Labs to conduct an in-depth evaluation ranging from standalone capture the flag(CTF) challenges to complex network environment simulations. The results reveal significant progress: Opus demonstrated markedly improved ability to think flexibly and adapt its approach to challenges instead of persisting with failed, unchanging approaches. Moreover, the model demonstrated significant improvement in vulnerability identification and executing complex multi-step attack chains, consistently succeeding where previous models failed. However, important limitations remain, particularly with maintaining coherent, long-horizon plans and goals if presented with unexpected obstacles. Our partners at Pattern Labs have posted the full evaluation report, which reveals both these exciting advances and critical limitations that inform our ongoing safety work.