# Improving Fairness in LLMs Through Testing-Time Adversaries

arXiv: https://arxiv.org/abs/2505.12100 · HTML: https://mlsanity.com/paper/2505.12100/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 26 references that are in the map.
- First posted on arXiv 17 May 2025.

## Builds on (most cited first, 20 of 20)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [Counterfactual Fairness](https://mlsanity.com/paper/1703.06856/) (2017)
- [Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distribution](https://mlsanity.com/paper/2202.10054/) (2022)
- [Bias and Fairness in Large Language Models: A Survey](https://mlsanity.com/paper/2309.00770/) (2023)
- [Larger language models do in-context learning differently](https://mlsanity.com/paper/2303.03846/) (2023)
- [Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations](https://mlsanity.com/paper/2204.02937/) (2022)
- [Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation](https://mlsanity.com/paper/1911.11834/) (2019)
- [Ask Me Anything: A simple strategy for prompting language models](https://mlsanity.com/paper/2210.02441/) (2022)
- [Capturing Failures of Large Language Models via Human Cognitive Biases](https://mlsanity.com/paper/2202.12299/) (2022)
- [SCALE-UP: An Efficient Black-box Input-level Backdoor Detection via Analyzing Scaled Prediction Consistency](https://mlsanity.com/paper/2302.03251/) (2023)
- [Large Language Models are Geographically Biased](https://mlsanity.com/paper/2402.02680/) (2024)
- [CyberSecEval 2: A Wide-Ranging Cybersecurity Evaluation Suite for Large Language Models](https://mlsanity.com/paper/2404.13161/) (2024)
- [On the Foundations of Shortcut Learning](https://mlsanity.com/paper/2310.16228/) (2023)
- [OpenBias: Open-set Bias Detection in Text-to-Image Generative Models](https://mlsanity.com/paper/2404.07990/) (2024)
- [Fairness in Large Language Models: A Taxonomic Survey](https://mlsanity.com/paper/2404.01349/) (2024)
- [Measuring Fairness in Generative Models](https://mlsanity.com/paper/2107.07754/) (2021)

## Abstract (arXiv)

Large Language Models (LLMs) push the bound-aries in natural language processing and generative AI, driving progress across various aspects of modern society. Unfortunately, the pervasive issue of bias in LLMs responses (i.e., predictions) poses a significant and open challenge, hindering their application in tasks involving ethical sensitivity and responsible decision-making. In this work, we propose a straightforward, user-friendly and practical method to mitigate such biases, enhancing the reliability and trustworthiness of LLMs. Our method creates multiple variations of a given sentence by modifying specific attributes and evaluates the corresponding prediction behavior compared to the original, unaltered, prediction/sentence. The idea behind this process is that critical ethical predictions often exhibit notable inconsistencies, indicating the presence of bias. Unlike previous approaches, our method relies solely on forward passes (i.e., testing-time adversaries), eliminating the need for training, fine-tuning, or prior knowledge of the training data distribution. Through extensive experiments on the popular Llama family, we demonstrate the effectiveness of our method in improving various fairness metrics, focusing on the reduction of disparities in how the model treats individuals from different racial groups. Specifically, using standard metrics, we improve the fairness in Llama3 in up to 27 percentage points. Overall, our approach significantly enhances fairness, equity, and reliability in LLM-generated results without parameter tuning or training data modifications, confirming its effectiveness in practical scenarios. We believe our work establishes an important step toward enabling the use of LLMs in tasks that require ethical considerations and responsible decision-making.