Critics challenge whether an AI ‘pain axis’ reflects felt pain

Dillon Plunkett and Robert Long are questioning claims around a “pain axis” found in language-model activations. Plunkett argues the work may track representations of a speaker in pain rather than a model’s own experience, and says emotional-pain prompts and random steering could explain some reported effects. The critics do not resolve whether AI can feel pain; they challenge what this experiment can show.