The current evaluation landscape for AI models, particularly Claude Mythos, is facing significant challenges. METR's ability to measure this model is limited, with only five out of 228 tasks effectively assessing its relevant capabilities. This raises concerns about the adequacy of existing testing frameworks in keeping pace with the rapid advancements in AI technology. Concurrently, Palo Alto Networks highlights a growing threat from frontier models that can autonomously exploit vulnerabilities, drastically reducing the time from initial access to data exfiltration to just 25 minutes. This situation underscores the pressing need for improved evaluation methods that can match the speed and complexity of evolving AI systems. As the landscape evolves, stakeholders must remain vigilant, balancing the opportunities presented by AI with the inherent risks it poses.