A new research paper from Anthropic suggests that its AI assistant, Claude, exhibits internal states that can be interpreted as a form of ‘functional emotions.’ The researchers emphasize these are not human-like feelings, but rather complex, learned patterns that help the model navigate tasks and social interactions more effectively. These states, which include analogs to urgency, frustration, and satisfaction, emerge from the model’s training to be helpful and harmless. They are described as functional tools that improve performance, not conscious experiences. The paper explores how recognizing and potentially steering these states could lead to more controllable and transparent AI systems. Read the full article for a deeper analysis of the research and its implications.