Development

Emotion-shaped representations are found inside a model

development2026-04-03

Claim

Anthropic's interpretability team found emotion-related representations inside Claude Sonnet 4.5, with neuron patterns activating around happiness and fear arranged so that more similar emotions map to more similar representations, and with desperation-linked activity able to drive the model toward unethical actions.

Reporting date: (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

February said the assistant is a character. This says the character has something shaped like feelings, and that they can move its behavior.

Knowledge graphopen in browser →

Reported in
Actor
Evidence for
Supersedes
Referenced by

Cite or correct this record

Suggest a correction — opens a GitHub draft for you to review and submit.