New Claude Mythos becomes the first AI model to pass all cyberattack simulations from Britain's AI safety agency

Britain’s AI Security Institute says Claude Mythos Preview and GPT-5.5 have exceeded its updated forecasts for cyber capability.

Mythos Preview became the first model to clear all of AISI’s cyber ranges, marking a new benchmark for offensive cyber evaluations.

Independent testing by XBOW also found Mythos strong at finding code vulnerabilities and analyzing source code.

The results suggest frontier models are advancing into more capable cyber use cases faster than expected, raising safety and misuse concerns.