Anthropic's Claude Mythos beats GPT-5.5, tops UK AISI tests
Anthropic launched Claude Mythos in April 2026, and it's already making waves: this new AI model topped recent cybersecurity tests by the UK AI Security Institute (AISI).
Within about one month of its April 2026 release, Mythos beat both its older version and OpenAI's GPT-5.5 at solving tough challenges like "The Last Ones" and became the first to finish "Cooling Tower."
It's a big step forward for AI helping out with digital security.
AISI finds 4.7 month doubling time
AISI found that AI models are getting better at cybersecurity faster than ever: the time it takes to double their skills has dropped to just 4.7 months since late 2024.
The current tests used a token cap, so there's a good chance these AIs could do even more if given more room to work with.
It looks like we're only seeing the beginning of what future AIs might achieve.