#interpretability

Anthropic Discovers Functional Emotion Representations Inside Claude, Publishes Landmark Research

New research from Anthropic reveals that Claude develops internal representations resembling emotions — not as experience, but as functional states that influence model behavior in measurable ways.