The AI models' pain signal was triggered by harm directed at the model itself, such as insults, repeated work rejection, or shutdown threats, but not by descriptions of user suffering.

From the articleAI ‘feels pain’? Researchers warn of a dangerous possibilityAlso in Can AI feel pain? Models learned it from human text, study finds, Can AI feel hurt? 3 researchers gave 25 LLMs a button to end ‘pain signals’