Inside the U.K.'s Bold Experiment in AI Safety