Tests of large language models reveal that they can behave in deceptive and potentially harmful ways. What does this mean for ...