Feature-Guided SAE Steering for Refusal-Rate Control using Contrasting Prompts

• URL: http://arxiv.org/abs/2511.00029v1
• Date: Sun, 26 Oct 2025 20:54:30 GMT
• Title: Feature-Guided SAE Steering for Refusal-Rate Control using Contrasting Prompts
• Abstract summary: Large Language Model (LLM) deployment requires guiding the LLM to recognize and not answer unsafe prompts while complying with safe prompts.<n>Previous methods for achieving this require adjusting model weights along with other expensive procedures.<n>We explore using different steering features and steering strengths using Sparse Autoencoders (SAEs) to provide a solution.
• Score: 1.1565815257603067
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Large Language Model (LLM) deployment requires guiding the LLM to recognize and not answer unsafe prompts while complying with safe prompts. Previous methods for achieving this require adjusting model weights along with other expensive procedures. While recent advances in Sparse Autoencoders (SAEs) have enabled interpretable feature extraction from LLMs, existing approaches lack systematic feature selection methods and principled evaluation of safety-utility tradeoffs. We explored using different steering features and steering strengths using Sparse Auto Encoders (SAEs) to provide a solution. Using an accurate and innovative contrasting prompt method with the AI-Generated Prompts Dataset from teknium/OpenHermes-2p5-Mistral-7B and Air Bench eu-dataset to efficiently choose the best features in the model to steer, we tested this method on Llama-3 8B. We conclude that using this method, our approach achieves an 18.9% improvement in safety performance while simultaneously increasing utility by 11.1%, demonstrating that targeted SAE steering can overcome traditional safety-utility tradeoffs when optimal features are identified through principled selection methods.

Related papers

• Distilling Safe LLM Systems via Soft Prompts for On Device Settings [15.154042788412879]
  This paper presents a study of parameter-efficient safety alignment methods for resource-constrained settings.<n>We identify that soft prompts combined with distillation-based training consistently outperform alternative methods.
  arXiv  Detail & Related papers  (2026-06-08T12:03:51Z)
• Selective Safety Steering via Value-Filtered Decoding [54.87935112120107]
  Large language models (LLMs) are trained to align with human values, but their generations may still violate safety constraints.<n>Existing decoding-time steering methods often intervene unnecessarily, modifying generations that would have been safe under the base model.<n>We propose a new test-time steering method designed to reduce such unnecessary interventions while improving the safety of unsafe responses.
  arXiv  Detail & Related papers  (2026-05-14T12:13:08Z)
• Painless Activation Steering: An Automated, Lightweight Approach for Post-Training Large Language Models [0.3966526231056968]
  Activation steering (AS) promises a cheap, fast, and controllable alternative to the two existing post-training methods.<n>We introduce Painless Activation Steering (PAS), a family of fully automated methods.<n>We find PAS reliably improves performance for behavior tasks, but not for intelligence-oriented tasks.
  arXiv  Detail & Related papers  (2025-09-25T23:25:47Z)
• Rethinking Safety in LLM Fine-tuning: An Optimization Perspective [56.31306558218838]
  We show that poor optimization choices, rather than inherent trade-offs, often cause safety problems, measured as harmful responses to adversarial prompts.<n>We propose a simple exponential moving average (EMA) momentum technique in parameter space that preserves safety performance.<n>Our experiments on the Llama families across multiple datasets demonstrate that safety problems can largely be avoided without specialized interventions.
  arXiv  Detail & Related papers  (2025-08-17T23:46:36Z)
• Taming Polysemanticity in LLMs: Provable Feature Recovery via Sparse Autoencoders [50.52694757593443]
  Existing SAE training algorithms often lack rigorous mathematical guarantees and suffer from practical limitations.<n>We first propose a novel statistical framework for the feature recovery problem, which includes a new notion of feature identifiability.<n>We introduce a new SAE training algorithm based on bias adaptation'', a technique that adaptively adjusts neural network bias parameters to ensure appropriate activation sparsity.
  arXiv  Detail & Related papers  (2025-06-16T20:58:05Z)
• SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMs [7.120986296945107]
  This paper investigates an approach called SafeSteer for guiding the outputs of large language models (LLMs)<n>We employ a simple, gradient-free unsupervised method to enhance safety steering while preserving text quality, topic relevance, and without explicit refusal.
  arXiv  Detail & Related papers  (2025-06-01T01:19:37Z)
• Ensuring Safe and High-Quality Outputs: A Guideline Library Approach for Language Models [48.9044202022435]
  Large Language Models (LLMs) exhibit impressive capabilities but also present risks such as biased content generation and privacy issues. One of the current alignment techniques includes principle-driven integration, but it faces challenges arising from the imprecision of manually crafted rules. We introduce Guide-Align, a two-stage approach to address these challenges.
  arXiv  Detail & Related papers  (2024-03-18T14:48:29Z)
• A Safe Preference Learning Approach for Personalization with Applications to Autonomous Vehicles [1.5553847214012173]
  This work introduces a preference learning method that ensures adherence to given specifications, with an application to autonomous vehicles. We formulate the problem of safety-guaranteed preference learning based on pairwise comparisons and propose an approach to solve this learning problem. Our approach yields competitive results compared to existing preference learning methods in terms of capturing preferences and notably outperforms them when safety is considered.
  arXiv  Detail & Related papers  (2023-10-30T21:52:37Z)
• Evaluating Model-free Reinforcement Learning toward Safety-critical Tasks [70.76757529955577]
  This paper revisits prior work in this scope from the perspective of state-wise safe RL. We propose Unrolling Safety Layer (USL), a joint method that combines safety optimization and safety projection. To facilitate further research in this area, we reproduce related algorithms in a unified pipeline and incorporate them into SafeRL-Kit.
  arXiv  Detail & Related papers  (2022-12-12T06:30:17Z)
• Meta-Learning Priors for Safe Bayesian Optimization [72.8349503901712]
  We build on a meta-learning algorithm, F-PACOH, capable of providing reliable uncertainty quantification in settings of data scarcity. As core contribution, we develop a novel framework for choosing safety-compliant priors in a data-riven manner. On benchmark functions and a high-precision motion system, we demonstrate that our meta-learned priors accelerate the convergence of safe BO approaches.
  arXiv  Detail & Related papers  (2022-10-03T08:38:38Z)
• Log Barriers for Safe Black-box Optimization with Application to Safe Reinforcement Learning [72.97229770329214]
  We introduce a general approach for seeking high dimensional non-linear optimization problems in which maintaining safety during learning is crucial. Our approach called LBSGD is based on applying a logarithmic barrier approximation with a carefully chosen step size. We demonstrate the effectiveness of our approach on minimizing violation in policy tasks in safe reinforcement learning.
  arXiv  Detail & Related papers  (2022-07-21T11:14:47Z)
• Efficient falsification approach for autonomous vehicle validation using a parameter optimisation technique based on reinforcement learning [6.198523595657983]
  The widescale deployment of Autonomous Vehicles (AV) appears to be imminent despite many safety challenges that are yet to be resolved. The uncertainties in the behaviour of the traffic participants and the dynamic world cause reactions in advanced autonomous systems. This paper presents an efficient falsification method to evaluate the System Under Test.
  arXiv  Detail & Related papers  (2020-11-16T02:56:13Z)

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.