New Claude Mythos becomes the first AI model to clear all cyberattack simulations from Britain's AI safety agency
Claude Mythos just became the first AI model to pass all of Britain's cyberattack simulations. The UK's safety agency has now revised its doubling timeline twice in months.
Why it matters
Anthropic's Claude Mythos has cleared a critical safety benchmark that no prior model achieved, signaling a sharp acceleration in AI cyber capabilities—and raising urgent questions about the pace of capability growth outstripping safety evaluation timelines.
The key facts
5 to knowClaude Mythos Preview is first model to clear all AISI cyberattack simulations
UK AI Security Institute revised cyber capability doubling timeline from 8 months to 4.7 months, then revised again
GPT-5.5 also surpassed accelerated timeline estimates
Anthropic's Logan Graham: 'Within a year, Mythos will probably look quite dumb'
Benchmark: AISI attack simulations (safety evaluation framework)
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: The UK's AI Security Institute has revised its estimate of how fast AI cyber capabilities are doubling—twice. First from eight months down to 4.7, and now Anthropic's Claude Mythos Preview and OpenAI's GPT-5.5 have blown past even that accelerated timeline. Mythos is the first model to clear all of…