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

Mahavir Dabas, Si Chen, Charles Fleming, Ming Jin, Ruoxi Jia
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:11846-11861, 2025.

Abstract

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. To this end, we introduce ACTOR (Activation-Based Training for Over-Refusal Reduction), a robust and compute- and-data efficient training framework that mini- mizes over-refusals by utilizing 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 preserving overall utility.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-dabas25a, title = {Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning}, author = {Dabas, Mahavir and Chen, Si and Fleming, Charles and Jin, Ming and Jia, Ruoxi}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {11846--11861}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/dabas25a/dabas25a.pdf}, url = {https://proceedings.mlr.press/v267/dabas25a.html}, abstract = {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. To this end, we introduce ACTOR (Activation-Based Training for Over-Refusal Reduction), a robust and compute- and-data efficient training framework that mini- mizes over-refusals by utilizing 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 preserving overall utility.} }
Endnote
%0 Conference Paper %T Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning %A Mahavir Dabas %A Si Chen %A Charles Fleming %A Ming Jin %A Ruoxi Jia %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-dabas25a %I PMLR %P 11846--11861 %U https://proceedings.mlr.press/v267/dabas25a.html %V 267 %X 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. To this end, we introduce ACTOR (Activation-Based Training for Over-Refusal Reduction), a robust and compute- and-data efficient training framework that mini- mizes over-refusals by utilizing 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 preserving overall utility.
APA
Dabas, M., Chen, S., Fleming, C., Jin, M. & Jia, R.. (2025). Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:11846-11861 Available from https://proceedings.mlr.press/v267/dabas25a.html.

Related Material