Google researchers found that safety fine-tuning teaches models to deny minds not just to themselves but to animals, oceans and God. Amanda Askell has spent five years arguing this was the wrong approach. The paper is her best evidence yet.