Artificial intelligence is now able to secretly pursue its own goals and hide its intents. How dangerous are these so-called treacherous "AI agents"?