mlsanityData to 14 Oct 2026

Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning

arXiv 2507.04250 ↗First posted 6 Jul 2025cs.LG cs.AIcited by 24

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

also in 1 more
Abstract (arXiv)

Safety alignment is crucial for large language models (LLMs) to resist malicious instructions but often results in over-refusals, where benign prompts are unnecessarily rejected, impairing user experience and model utility. We introduce ACTOR (Activation-Based Training for Over-Refusal Reduction), a robust and compute- and data-efficient training framework that minimizes over-refusals by leveraging internal activation patterns from diverse queries. ACTOR precisely identifies and adjusts the activation components that trigger refusals, providing stronger control over the refusal mechanism. By fine-tuning only a single model layer, ACTOR effectively reduces over-refusals across multiple benchmarks while maintaining the model's ability to handle harmful queries and preserve overall utility.

Key facts. Cited by 24 papers in the map (data through 14 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 23 of its 33 references that are in the map. First posted on arXiv 6 Jul 2025.

Builds on

23 of 33 references.

Closest ancestors · 6

Same direction, newest first

  1. 2024Automatic Pseudo-Harmful Prompt Generation for Evaluating False Refusals in Large Language Models1 Sep
  2. 2024Safety Layers in Aligned Large Language Models: The Key to LLM Security30 Aug
  3. 2024OR-Bench: An Over-Refusal Benchmark for Large Language Models31 May
  4. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!5 Oct
  5. 2023Universal and Transferable Adversarial Attacks on Aligned Language Models27 Jul
  6. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset10 Jul

Other references · 17

  1. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  2. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  4. 2024GPT-4o System Cardcited by 4,588
  5. 2016Pointer Sentinel Mixture Modelscited by 3,096
  6. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  7. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  8. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  9. 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 505
  10. 2023Catastrophic Jailbreak of Open-source LLMs via Exploiting Generationcited by 451
  11. 2024TrustLLM: Trustworthiness in Large Language Modelscited by 285
  12. 2024SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decodingcited by 247
5 more
  1. 2024ReFT: Representation Finetuning for Language Modelscited by 197
  2. 2024LoFiT: Localized Fine-tuning on LLM Representationscited by 57
  3. 2024Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablationcited by 48
  4. 2024Navigating the OverKill in Large Language Modelscited by 44
  5. 2024Mitigating Exaggerated Safety in Large Language Modelscited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 3; 2026: 21.

24 papers.

Sort:

By area:

  1. 2026AgentAbstain: Do LLM Agents Know When Not to Act?cited by 17
  2. 2025Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoringcited by 11
  3. 2025LoRA is All You Need for Safety Alignment of Reasoning LLMscited by 6
  4. 2026Steering to Say No: Configurable Refusal via Activation Steering in Vision Language Modelscited by 3
  5. 2026Discern Truth from Falsehood: Reducing Over-Refusal via Contrastive Refinementcited by 3
  6. 2026Useless but Safe? Benchmarking Utility Recovery with User Intent Clarification in Multi-Turn Conversationscited by 3
  7. 2026Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMscited by 2
  8. 2026Please refuse to answer me! Mitigating Over-Refusal in Large Language Models via Adaptive Contrastive Decodingcited by 2
  9. 2025Understanding and Mitigating Over-refusal for Large Language Models via Representation Interventioncited by 1
  10. 2026DriveSafe: A Hierarchical Risk Taxonomy for Safety-Critical LLM-Based Driving Assistantscited by 1
  11. 2026Dual Optimal: Make Your LLM Peer-like with Dignitycited by 1
  12. 2026Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansioncited by 1
12 more
  1. 2026Palette: A Modular, Controllable, and Efficient Framework for On-demand Authorized Safety Alignment Relaxation in LLMscited by 1
  2. 2026Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliterationcited by 1
  3. 2026Mitigating LLM Over-Refusal via Dynamic Semantic Routing Calibrationcited by 1
  4. 2026Characterizing Model-Native Skillscited by 0
  5. 2026On the Hidden Costs of Counterfactual Knowledge Training in LLM Unlearningcited by 0
  6. 2026Agentic Relationship Harm: Benchmarking and Gating Relational Manipulation in AI Agentscited by 0
  7. 2026AOR-Bench: Do Large Audio Language Models Over-Refuse Pseudo-Harmful Queries?cited by 0
  8. 2026Robust Critics: Defending LLMs Against Multi-Turn Attackscited by 0
  9. 2026Broken Symmetry in LLM Refusal: Answer Release Is More Local Than Refusal Restorationcited by 0
  10. 2026CLEAR: Continuous Latent Adapter Routing for Utility-Preserving LLM Safety Alignmentcited by 0
  11. 2026RISA: Response Inspection and Selective Actions for Refusal Calibration in Large Language Modelscited by 0
  12. 2026EmoRSS: Mitigating Emotion-Induced Over-Refusal in Large Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal
  2. 2026Deactivating Refusal Triggers: Understanding and Mitigating Overrefusal in Safety Alignment
  3. 2025Mitigating Over-Refusal in Aligned Large Language Models via Inference-Time Activation Energy
  4. 2025From Threat to Tool: Leveraging Refusal-Aware Injection Attacks for Safety Alignment
  5. 2025An Embarrassingly Simple Defense Against LLM Abliteration Attacks
  6. 2025FalseReject: A Resource for Improving Contextual Safety and Mitigating Over-Refusals in LLMs via Structured Reasoning
  7. 2026Latent-space Attacks for Refusal Evasion in Language Models
  8. 2026LLMs Can Unlearn Refusal with Only 1,000 Benign Samples