SOURCE-LINKED INTELLIGENCE

Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering

arXiv · AI, language, vision and robotics · article · Aug 31, 2026 · UTC

Activation steering has emerged as a lightweight approach for controlling model refusal at inference time. A growing line of research explores trainable rotations of activations to develop geometrically principled intervention mechanisms. However, existing techniques rely on auxiliary constructs, such as refusal vectors, to define these rotations. In our work, we develop a self-contained methodology for learning parameter-efficient rotational transformations based on Riemannian optimization. We empirically validate the proposed scheme, demonstrating its superiority in intervention efficiency.

Read original source ↗ Open in workspace

recordType

paper

region

Global