![[2512.16602] Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2512.16602] Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics
Summary
The paper introduces Refusal Steering, a method for controlling Large Language Models' refusal behavior on sensitive topics without retraining, enhancing moderation capabilities.
Why It Matters
As AI systems increasingly engage with sensitive topics, ensuring appropriate responses is crucial. Refusal Steering offers a solution to manage refusal behavior effectively, maintaining safety while allowing for nuanced interactions. This method could improve AI transparency and user trust in sensitive discussions.
Key Takeaways
- Refusal Steering allows fine-grained control over LLM refusal behavior.
- The method does not require retraining and can be applied at inference time.
- It maintains safety while removing refusals on politically sensitive topics.
- The approach generalizes across different model sizes (4B and 80B).
- Activation steering can help achieve targeted refusals when necessary.
Computer Science > Computation and Language arXiv:2512.16602 (cs) [Submitted on 18 Dec 2025 (v1), last revised 24 Feb 2026 (this version, v3)] Title:Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics Authors:Iker García-Ferrero, David Montero, Roman Orus View a PDF of the paper titled Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics, by Iker Garc\'ia-Ferrero and 2 other authors View PDF HTML (experimental) 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...