Anthropic's Claude Mythos sets records at UK A.I. security institute
Technology
Anthropic's Claude Mythos just set new records in cybersecurity tests run by the U.K.'s A.I. Security Institute.
It solved tricky challenges like The Last Ones and Cooling Tower, tasks that Anthropic's newer Mythos Preview checkpoint outperformed OpenAI's GPT-5.5 on, showing how fast these models are improving.
Experts urge better model measurement
The institute says the tests had a 2.5 million-token cap, so Mythos might be even more powerful than results suggest.
With A.I. tech advancing faster every few months, experts say it's time for better ways to measure what these models can really do, especially as they start tackling bigger real-world problems.