ai Refusal in Language Models Is Mediated by a Single Direction You can delete a model's "no" with linear algebra. Safety refusal ≈ one vector. Remove it. Model complies. Abliteration on Qwen3.8-27B: 5h A100.
