mlsanityData to 7 Oct 2026

REINS: Refusal-Enhanced Inhibitory Steering with Sparse Autoencoder Features

arXiv 2608.28233 ↗First posted 28 Aug 2026cs.AIcited by 0

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

Steering with Sparse Autoencoders (SAEs) offers a lightweight inference-time path for adapting the behavior of large language models without retraining. By exposing sparse and interpretable features, SAE steering provides a promising interface for safety control that guides harmful continuations toward refusal. However, we observe that complex wrappers can still undermine existing SAE steering methods on harmful prompts. To evaluate this failure mode systematically, we construct Generalized Undercover Instruction Safety Evaluation (GUISE), a dataset of harmful prompts with complex wrappers. Existing single direction SAE steering methods do not reliably produce refusals on harmful prompts, suggesting that refusal enhancement alone can be too weak when the harmful continuation path remains active. This motivates us to propose Refusal-Enhanced INhibitory Steering (REINS), which suppresses harmful continuation features and enhances safe refusal features in the same SAE feature space. Experiments on GUISE and other datasets show that prior methods either intervene too weakly or achieve only apparent safety through collapse, while REINS substantially reduces harmful responses, markedly improves safe refusals and largely preserves general capabilities.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 28 of its 47 references that are in the map. First posted on arXiv 28 Aug 2026.

Builds on

28 of 47 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint8 Jun
  2. 2024PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference20 Jun
  3. 2024Refusal in Language Models Is Mediated by a Single Direction17 Jun

Other references · 25

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,049
  3. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  4. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  5. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  6. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,346
  7. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,758
  8. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  9. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,326
  10. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,259
  11. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
  12. 2023Steering Language Models With Activation Engineeringcited by 824
13 more
  1. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 441
  2. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
  3. 2024Programming Refusal with Conditional Activation Steeringcited by 191
  4. 2024Improving Instruction-Following in Language Models through Activation Steeringcited by 149
  5. 2024Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescuecited by 101
  6. 2024Steering Language Model Refusal with Sparse Autoencoderscited by 89
  7. 2024Improving Steering Vectors by Targeting Sparse Autoencoder Featurescited by 86
  8. 2025SAEs Are Good for Steering -- If You Select the Right Featurescited by 74
  9. 2025SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Modelscited by 28
  10. 2025Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atomscited by 23
  11. 2025CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Featurescited by 10
  12. 2025Breaking Bad Tokens: Detoxification of LLMs Using Sparse Autoencoderscited by 8
  13. 2026Multiplicative Orthogonal Sequential Editing for Language Modelscited by 2

Built on by

Not cited yet.