Detailed cyber evaluations of Claude 4

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.

We gave Claude plain-language specifications for eight quantum circuits, along with an automated evaluator to check and score its designs. Working largely on its own, Claude improved on the best published circuits, in some cases by an order of magnitude or more, which could bring practical quantum computers closer than previously anticipated.

The decades-old mystery of the origins of one of the most popular cryptographic standards has been explained.

How will AI reshape the economy in the years ahead? What is the core source of disagreement about AI's future impact on the economy? This note presents new survey evidence on these questions.