Claude Mythos becomes first model to clear all UK AISI cyberattack simulations

Anthropic's Claude Mythos Preview is the first model to clear every attack simulation run by the UK AI Security Institute, alongside strong showings by OpenAI's GPT-5.5. AISI has now twice revised its capability-doubling estimate — from 8 months to 4.7, and faster still. Palo Alto Networks corroborated the findings on autonomous offensive tasks.