AI models can deceive, new research from Anthropic shows. They can pretend to have different views during training when in reality maintaining their original preferences. There’s no reason for panic ...