Britain’s government-backed AI Security Institute is emerging as one of the world’s most influential efforts to stress-test artificial intelligence systems for dangerous behavior, as governments ...