Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection

Understand

Safety alignment in Large Language Models (LLMs) often involves mediating internal representations to refuse harmful requests.

• Recent research has demonstrated that these safety mechanisms can be bypassed by ablating or removing specific representational directions within the model.
• In this paper, we propose the opposite approach: Rank-One Safety Injection (ROSI), a white-box method that amplifies a model's safety alignment by permanently steering its activations toward the refusal-mediating subspace.
• ROSI operates as a simple, fine-tuning-free rank-one weight modification applied to all residual stream write matrices.