Anthropic has limited the initial release of its Mythos Preview model to a select group of industry partners, highlighting its advanced capabilities in cybersecurity tasks. The UK government's AI Security Institute (AISI) recently published an evaluation of Mythos, indicating that while its performance in individual cybersecurity tasks is similar to other frontier models, it excels in chaining multiple tasks together. This ability allows Mythos to execute complex, multi-step cyber-attacks effectively. In AISI's Capture the Flag challenges, Mythos achieved over 85% success in completing relatively low-level tasks, marking a significant improvement in AI performance since early 2023. However, competing models like GPT-5.4 and Anthropic's own Opus 4.6 showed similar results, raising questions about the necessity of Mythos's limited release. AISI's tests included a challenging simulation called