Two-faced AI language models learn to hide deception

Open in new window