mlsanityData to 14 Oct 2026

Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering

arXiv 2608.30986 ↗First posted 31 Aug 2026cs.LG cs.CLcited by 0

On the map: Mechanistic interpretability → Steering and probing for LLM safety

Abstract (arXiv)

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. An extensive ablation study highlights the importance of key design choices in our method. Our results identify the proposed rotation-based steering scheme as a promising direction for more reliable control over the behavior of LLMs.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Steering and probing for LLM safety, one of 13 directions in Mechanistic interpretability. Builds on 25 of its 31 references that are in the map. First posted on arXiv 31 Aug 2026.

Builds on

25 of 31 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal7 Sep
  2. 2025AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint8 Jun

Other references · 23

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  3. 2024Qwen2.5 Technical Reportcited by 4,942
  4. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,886
  5. 2016Pointer Sentinel Mixture Modelscited by 3,096
  6. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,207
  7. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
  8. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  9. 2024DeepSeek LLM: Scaling Open-Source Language Models with Longtermismcited by 642
  10. 2021Editing Factual Knowledge in Language Modelscited by 553
  11. 2024SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Modelscited by 228
  12. 2025Qwen3Guard Technical Reportcited by 155
11 more
  1. 2024Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimizationcited by 118
  2. 2025The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independencecited by 115
  3. 2025LLMs Encode Harmfulness and Refusal Separatelycited by 88
  4. 2023Eight Things to Know about Large Language Modelscited by 73
  5. 2025Angular Steering: Behavior Control via Rotation in Activation Spacecited by 53
  6. 2024Refusal in LLMs is an Affine Functioncited by 29
  7. 2024Householder Pseudo-Rotation: A Novel Approach to Activation Editing in LLMs with Direction-Magnitude Perspectivecited by 23
  8. 2026Spherical Steering: Geometry-Aware Activation Rotation for Language Modelscited by 21
  9. 2025SOM Directions are Better than One: Multi-Directional Refusal Suppression in Language Modelscited by 9
  10. 2026Selective Steering: Norm-Preserving Control Through Discriminative Layer Selectioncited by 6
  11. 2026Minimizing Collateral Damage in Activation Steeringcited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Universal Refusal Circuits Across LLMs: Cross-Model Transfer via Trajectory Replay and Concept-Basis Reconstruction
  2. 2026Local Linearity of LLMs Enables Activation Steering via Model-Based Linear Optimal Control
  3. 2025Adaptive Activation Steering for Efficient LLM Reasoning via Closed-Loop PID Control
  4. 2026Steering to Say No: Configurable Refusal via Activation Steering in Vision Language Models
  5. 2026On the Steering Dimensionality of Refusal in Language Models
  6. 2026C-DeltaTheta: Circuit-Restricted Weight Arithmetic for Selective Refusal
  7. 2024Programming Refusal with Conditional Activation Steering
  8. 2025Guiding Giants: Lightweight Controllers for Weighted Activation Steering in LLMs