UK gov's Mythos AI tests help separate cybersecurity threat from hype

AISI's evaluation found the Mythos Preview model significantly outperformed prior models on a simulated multi-step intrusion benchmark (TLO), completing many more infiltration steps and demonstrating the potential to autonomously attack small, weakly defended enterprise systems; however, the tests are constrained by simulated vulnerabilities, limited compute budget, and absence of active defenders, and Mythos still struggled with harder scenarios, so its real-world effectiveness against well-defended systems remains uncertain.