Wait, am I Being Fair? Characterizing Deductive Stereotyping and Mitigating It with Fair-GCG
Summary
This paper identifies deductive stereotyping in LLMs and proposes Fair-GCG, a reasoning-time injection framework to mitigate biased inferences in fairness tasks.
View Cached Full Text
Cached at: 07/01/26, 05:31 AM
# Wait, am I Being Fair? Characterizing Deductive Stereotyping and Mitigating It with Fair-GCG Source: [https://arxiv.org/abs/2606.30989](https://arxiv.org/abs/2606.30989) [View PDF](https://arxiv.org/pdf/2606.30989) > Abstract:Warning: This paper contains several toxic and offensive statements\. While reasoning generally improves fairness in recent large language models \(LLMs\), failures persist\. In this work, we identify a failure mode, deductive stereotyping, in which models apply population\-level statistical regularities to individual cases, producing logically coherent yet socially biased inferences\. We provide a statistical interpretation of this phenomenon\. To steer models toward fairness\-aware reasoning, we propose a reasoning\-time injection framework\. We further introduce Fair\-GCG to systematically discover effective injection phrases\. Injection phrases discovered by Fair\-GCG improve performance across multiple fairness benchmarks, generalize from smaller to larger LLMs, improves reasoning\-level fairness, reduces bias in open\-ended generation, and transfer to real\-world fairness\-sensitive tasks\. ## Submission history From: Naihao Deng \[[view email](https://arxiv.org/show-email/1edcc79b/2606.30989)\] **\[v1\]**Tue, 30 Jun 2026 00:00:42 UTC \(1,143 KB\)
Similar Articles
Detecting and Mitigating Bias by Treating Fairness as a Symmetry Operation
The paper proposes treating fairness as a symmetry operation in machine learning classifiers, implementing loss-based regularization to enforce invariance under swapping of sensitive attributes while holding merit features fixed. The framework achieves over 90% bias reduction with minimal accuracy loss and requires no causal graph knowledge.
Fair outputs, Biased Internals: Causal Potency and Asymmetry of Latent Bias in LLMs for High-Stakes Decisions
This paper studies how instruction-tuned LLMs can exhibit fair outputs while retaining biased internal representations in high-stakes decisions like mortgage underwriting, showing that these hidden biases are causally potent, asymmetric, and exploitable through activation steering.
Do Counterfactually Fair Image Classifiers Satisfy Group Fairness? -- A Theoretical and Empirical Study
This paper theoretically and empirically studies the relationship between counterfactual fairness (CF) and group fairness (GF) in image classification, introducing new CF evaluation datasets (CelebA-CF and LFW-CF). It finds that CF does not imply GF in images due to latent attributes correlated with sensitive attributes, and proposes Counterfactual Knowledge Distillation (CKD) to mitigate this.
Explanation Fairness in Large Language Models: An Empirical Analysis of Disparities in How LLMs Justify Decisions Across Demographic Groups
This paper introduces the Explanation Fairness Taxonomy (EFT) to analyze disparities in how LLMs justify decisions across demographic groups, finding significant biases in explanation quality and tone despite balanced decisions.
The Geopolitics of AI Safety: A Causal Analysis of Regional LLM Bias
This paper introduces a Probabilistic Graphical Model framework to causally audit LLM safety mechanisms, revealing that standard observational metrics overestimate demographic bias by ignoring context toxicity.