How Does Differential Privacy Affect Social Bias in LLMs? A Systematic Evaluation
Quick Take
Differential privacy reduces bias in some LLM tasks but not universally across all paradigms.
Key Points
- DP-SGD trained LLMs show varied bias reduction.
- Bias improvement is task-specific and not universal.
- Multi-paradigm evaluation is crucial for fairness assessment.
📖 Reader Mode
~2 min readAbstract:Large language models (LLMs) trained on web-scale corpora can memorize sensitive training data, posing significant privacy risks. Differential privacy (DP) has emerged as a principled framework that limits the influence of individual data points during training, yet the relationship between differential privacy and social bias in LLMs remains poorly understood. To investigate this, we present a systematic evaluation of social bias in a pretrained LLM trained with DP-SGD, comparing a DP model against non-DP baselines across four complementary paradigms: sentence scoring, text completion, tabular classification, and question answering. We find that DP reduces bias in sentence scoring tasks, where bias is measured through controlled likelihood comparisons, yet this improvement does not generalize across all tasks. Our results reveal a discrepancy between logit-level bias and output-level bias. Moreover, decreasing memorization does not necessarily reduce unfairness, underscoring the importance of multi-paradigm evaluation when assessing fairness in LLMs.
| Comments: | 14 pages, 1 figure |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.11195 [cs.CL] |
| (or arXiv:2605.11195v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.11195 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Eduardo Tenorio [view email]
[v1]
Mon, 11 May 2026 20:03:05 UTC (129 KB)
— Originally published at arxiv.org
More from arXiv cs.CL
See more →arXiv cs.CL·Luis Lara, Aristides Milios, Zhi Hao Luo, Aditya Sharma, Ge Ya Luo, Christopher Beckham, Florian Golemo, Christopher Pal
2d ago FeaturedOriginal
Generative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards
AI Summary
A new LLM-based approach generates floor plans while adhering to numerical and topological constraints using reinforcement learning.