Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering

• URL: http://arxiv.org/abs/2608.30986v1
• Date: Mon, 31 Aug 2026 15:43:47 GMT
• Title: Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering
• Abstract summary: We develop a self-contained methodology for learning parameter-efficient rotational transformations.<n>We empirically validate the proposed scheme, demonstrating its superiority in intervention efficiency.<n>Our results identify the proposed rotation-based steering scheme as a promising direction for more reliable control over the behavior of LLMs.
• Score: 41.986377298239525
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• Retraction-Free Optimization over the Stiefel Manifold for the LoRA Fine-Tuning [17.71677083154088]
  datasets over the Stiefel play a significant role in various machine learning tasks.<n>Existing methods either use the manifold operators, requiring costly orthonormalization for large-scale matrices, or employ manifold tuning methods.
  arXiv  Detail & Related papers  (2026-07-28T05:22:39Z)
• Activation Steering Induces Emergent Misalignment: A More Comprehensive Evaluation [74.17379276939599]
  We show that activation steering can induce broad misalignment, even in the recent Qwen-3.5 series.<n>We characterize properties of AS-induced EM by analyzing key steering-specific factors, including steering magnitude, the low-rank structure of the steering subspace, and the number of epochs during steering-vector construction.
  arXiv  Detail & Related papers  (2026-06-07T15:34:59Z)
• Learning Structured Reasoning via Tractable Trajectory Control [99.75278337895024]
  Ctrl-R is a framework for learning structured reasoning via tractable trajectory control.<n>We show that Ctrl-R enables effective exploration and internalization of previously unattainable reasoning patterns.
  arXiv  Detail & Related papers  (2026-03-02T09:18:19Z)
• ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment [49.68063561145927]
  We propose a unified ordinary differential equations (ODEs)-based theoretical framework for activation steering.<n>We introduce ODESteer, a kind of ODE-based steering guided by barrier functions.<n>Compared to state-of-the-art activation steering methods, ODESteer achieves consistent empirical improvements.
  arXiv  Detail & Related papers  (2026-02-19T17:13:44Z)
• ODELoRA: Training Low-Rank Adaptation by Solving Ordinary Differential Equations [54.886931928255564]
  Low-rank adaptation (LoRA) has emerged as a widely adopted parameter-efficient fine-tuning method in deep transfer learning.<n>We propose a novel continuous-time optimization dynamic for LoRA factor matrices in the form of an ordinary differential equation (ODE)<n>We show that ODELoRA achieves stable feature learning, a property that is crucial for training deep neural networks at different scales of problem dimensionality.
  arXiv  Detail & Related papers  (2026-02-07T10:19:36Z)
• Equivariant Goal Conditioned Contrastive Reinforcement Learning [5.019456977535218]
  Contrastive Reinforcement Learning (CRL) provides a promising framework for extracting useful structured representations from unlabeled interactions.<n>We propose Equivariant CRL, which further structures the latent space using equivariant constraints.<n>Our approach consistently outperforms strong baselines across a range of simulated tasks in both state-based and image-based settings.
  arXiv  Detail & Related papers  (2025-07-22T01:13:45Z)
• Householder Pseudo-Rotation: A Novel Approach to Activation Editing in LLMs with Direction-Magnitude Perspective [32.12158544745983]
  We propose a novel editing method that views activations in terms of their directions and magnitudes.<n>Our method, named Householder Pseudo-Rotation (HPR), mimics the rotation transformation, thus preserving activation norms and resulting in an improved performance on various safety benchmarks.
  arXiv  Detail & Related papers  (2024-09-16T07:29:40Z)
• ASR: Attention-alike Structural Re-parameterization [53.019657810468026]
  We propose a simple-yet-effective attention-alike structural re- parameterization (ASR) that allows us to achieve SRP for a given network while enjoying the effectiveness of the attention mechanism. In this paper, we conduct extensive experiments from a statistical perspective and discover an interesting phenomenon Stripe Observation, which reveals that channel attention values quickly approach some constant vectors during training.
  arXiv  Detail & Related papers  (2023-04-13T08:52:34Z)
• A Data-Driven Model-Reference Adaptive Control Approach Based on Reinforcement Learning [4.817429789586126]
  A model-reference adaptive solution is developed here for autonomous systems where it solves the Hamilton-Jacobi-Bellman equation of an error-based structure. This is done in real-time without knowing or employing the dynamics of either the process or reference model in the control strategies.
  arXiv  Detail & Related papers  (2023-03-17T14:10:52Z)
• Reinforcement Learning for Low-Thrust Trajectory Design of Interplanetary Missions [77.34726150561087]
  This paper investigates the use of reinforcement learning for the robust design of interplanetary trajectories in presence of severe disturbances. An open-source implementation of the state-of-the-art algorithm Proximal Policy Optimization is adopted. The resulting Guidance and Control Network provides both a robust nominal trajectory and the associated closed-loop guidance law.
  arXiv  Detail & Related papers  (2020-08-19T15:22:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.