What prevents a rogue ai agent from identifying as belonging to a competing ai lab?