THE AI Security Institute (AISI) has urged organisations to double down on cybersecurity basics and consider harnessing AI to protect systems after testing Anthropic’s Mythos Preview. According to the AI Security Institute, Mythos Preview could execute multi-stage attacks on vulnerable networks and discover and exploit vulnerabilities autonomously, tasks that would take human professionals days of work.
In its evaluation, the institute built a 32-step corporate network attack simulation; Mythos solved it in only three out of ten attempts, and completed 22 of 32 steps on average across all attempts, suggesting that more inference compute might improve performance. However, the AISI noted that its testing environment differs in important ways from real-world environments, lacking active defenders and defensive tooling and imposing no penalties for actions that would trigger security alerts.
In the meantime, it urged security teams to strengthen baseline protections and to consider AI to help deliver improvements in defence, while planning to simulate defended environments with endpoint detection and real-time incident response in the future.