OPIUM: Mitigating Steering Externalities and Over-Refusal via Dual Objective Latent Optimization
On the map: LLM safety and security → Safety alignment, refusal and guardrails
also in 1 more
Abstract (arXiv)
Activation steering provides a lightweight mechanism for controlling large language models at inference time, but steering vectors can have unintended externalities: utility vectors may weaken safety behavior, while refusal vectors may induce over-refusal on benign prompts. We introduce OPIUM (Optimizing Protected Injections via Utility Manifolds), a training-free method for sanitizing steering vectors through representation matching. Given reference behaviors on two prompt sets, OPIUM optimizes a new steering vector that preserves the downstream representations induced by the desired intervention while matching a safer reference behavior on prompts where the original vector fails. Across steering-externality and over-refusal settings, OPIUM improves the safety--utility tradeoff relative to vanilla steering and directional ablation, suggesting that harmful side effects of activation steering can often be mitigated directly in activation space.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 18 of its 21 references that are in the map. First posted on arXiv 22 Jul 2026.
Builds on
18 of 21 references.
Closest ancestors · 7
Same direction, newest first
- 2026Steering Externalities: Benign Activation Steering Unintentionally Increases Jailbreak Risk for Large Language Models3 Feb
- 2025The Rogue Scalpel: Activation Steering Compromises LLM Safety26 Sep
- 2025FalseReject: A Resource for Improving Contextual Safety and Mitigating Over-Refusals in LLMs via Structured Reasoning12 May
- 2024Who's asking? User personas and the mechanics of latent misalignment17 Jun
- 2024Refusal in Language Models Is Mediated by a Single Direction17 Jun
- 2024Safety Alignment Should Be Made More Than Just a Few Tokens Deep10 Jun
- 2024OR-Bench: An Over-Refusal Benchmark for Large Language Models31 May
Other references · 11
- 2024The Llama 3 Herd of Modelscited by 14,715
- 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
- 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,420
- 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
- 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
- 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
- 2023Steering Language Models With Activation Engineeringcited by 922
- 2023Tree of Attacks: Jailbreaking Black-Box LLMs Automaticallycited by 672
- 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 505
- 2024Programming Refusal with Conditional Activation Steeringcited by 206
- 2024Extending Activation Steering to Broad Skills and Multiple Behaviourscited by 40
Built on by
Citing papers per year: 2026: 2.
2 papers.
Similar, not cited
Close in wording, no citation either way
- 2026On the Steering Dimensionality of Refusal in Language Models
- 2026REINS: Refusal-Enhanced Inhibitory Steering with Sparse Autoencoder Features
- 2024Steering Language Model Refusal with Sparse Autoencoders
- 2026From Refusal Tokens to Refusal Control: Discovering and Steering Category-Specific Refusal Directions
- 2026What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal
- 2026Steer Like the LLM: Activation Steering that Mimics Prompting
- 2025Understanding (Un)Reliability of Steering Vectors in Language Models
- 2025Improved Representation Steering for Language Models