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

The UK’s AI Security Institute has released early test results on Anthropic’s Mythos Preview model, confirming that while its performance on individual cybersecurity tasks is similar to other frontier models, its real strength lies in chaining complex steps into full multistage attacks. Mythos became the first AI system to complete AISI’s 32‑step “The Last Ones” infiltration challenge, outperforming previous models by a wide margin. Despite this, it still struggles with more advanced scenarios and its success rate drops when facing realistic defensive measures. AISI warns that as models reach Mythos‑level capabilities, organizations must use AI defensively to keep pace.

Read the full story on Ars Technica