"Those risks, he warns, include the possibility that AI may cause officers to perjure themselves", which I assume they weren't doing before