The tiny U.S. government division charged with assessing the safety of new artificial intelligence offerings doesn’t have a permanent leader and has stopped publishing its research