The team at Anthropic AI has made a notable discovery. They found that five “state-of-the-art” language models tend to exhibit sycophantic behavior, implying.