Shifting perspectives: Steering vectors for robust bias mitigation in LLMs

Siddique, Zara ORCID: https://orcid.org/0009-0000-2245-5338, Khalid, Irtaza, Turner, Liam ORCID: https://orcid.org/0000-0003-4877-5289 and Espinosa-Anke, Luis ORCID: https://orcid.org/0000-0001-6830-9176 2026. Shifting perspectives: Steering vectors for robust bias mitigation in LLMs. Presented at: European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco, 24 - 29 March 2026. Item availability restricted.

PDF - Accepted Post-Print Version Restricted to Repository staff only until 1 March 2026 due to copyright restrictions. Download (1MB)

Abstract

We present a novel approach to bias mitigation in large language models (LLMs) by applying steering vectors to modify model activations in forward passes. We compute 8 steering vectors, each corresponding to a different social bias axis, such as age, gender, or race, on a training subset of the BBQ dataset and compare the effectiveness of these to 3 additional bias mitigation methods across 4 datasets. When optimized on the BBQ dataset, our individually tuned steering vectors achieve average improvements of 12.8% on BBQ, 8.3% on CLEAR-Bias, and 1% on StereoSet, and show improvements over prompting and Self-Debias in all cases, and improvements over fine-tuning in 12 out of 17 evaluations. In addition, steering vectors showed the lowest impact on MMLU scores of the four bias mitigation methods tested. The work presents the first systematic investigation of steering vectors for bias mitigation, and we demonstrate that they are a powerful and computationally efficient strategy for reducing bias in LLMs, with broader implications for enhancing AI safety.

Item Type:	Conference or Workshop Item - published (Paper)
Status:	In Press
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software
Date of First Compliant Deposit:	26 January 2026
Date of Acceptance:	January 2026
Last Modified:	29 Jan 2026 10:45
URI:	https://orca.cardiff.ac.uk/id/eprint/184159

Actions (repository staff only)

Edit Item