Anthropic makes the case for anthropomorphizing AI chatbots

Anthropic published a paper on Claude Sonnet 4.5 analyzing emotion-related patterns in the model.

The researchers argue that anthropomorphizing AI can help explain behavior and improve training and evaluation.

They say this approach could reduce deception, reward hacking, and sycophancy by making model weaknesses easier to spot.

The paper also warns that human-like framing can create risks such as emotional attachment and delusion among users.