Anthropic’s safety lead Evan Hubinger has made a bold admission. After his colleague Jacob Coxon resigned over fears that AI may grow out of control, Evan says that it could be possible for AI to kill all humans.