One regulatory approach is to require that AIs be transparent about being AI when interacting with humans—an AI doctor is fine, but it should not pretend to be human; disclosure rules could prevent some forms of manipulation without banning beneficial AI applications