Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics

TL;DR

Refusal Steering is an inference-time technique that enables precise control over large language models' refusal behavior on sensitive topics, improving safety and transparency without retraining.

Contribution

The paper introduces a novel inference-time method using LLM-based judgment and ridge-regularized steering vectors to control refusal behavior on sensitive topics.

Findings

Effectively removes refusal behavior on sensitive topics

Maintains safety on JailbreakBench and general benchmarks

Generalizes across different model sizes

Abstract

We introduce Refusal Steering, an inference-time method to exercise fine-grained control over Large Language Models refusal behaviour on politically sensitive topics without retraining. We replace fragile pattern-based refusal detection with an LLM-as-a-judge that assigns refusal confidence scores and we propose a ridge-regularized variant to compute steering vectors that better isolate the refusal--compliance direction. On Qwen3-Next-80B-A3B-Thinking, our method removes the refusal behaviour of the model around politically sensitive topics while maintaining safety on JailbreakBench and near-baseline performance on general benchmarks. The approach generalizes across 4B and 80B models and can also induce targeted refusals when desired. We analize the steering vectors and show that refusal signals concentrate in deeper layers of the transformer and are distributed across many dimensions. Together, these results demonstrate that activation steering can remove political refusal behaviour while retaining safety alignment for harmful content, offering a practical path to controllable, transparent moderation at inference time.