Do LLMs Really Lack Difference Awareness? Uncovering Heuristic Substitution in Group-Differentiating Fairness

Large language models (LLMs) tuned for safety often avoid acknowledging demographic differences, even when such acknowledgment is factually correct (e.g., ancestry-based disease incidence) or contextually justified (e.g., religious hiring preferences). This identity-blindness yields incorrect responses, unnecessary refusals, or generic "equal-treatment" defaults. We study this via difference-awareness classification: given a question involving demographic groups, the task is not to answer directly, but to classify whether a correct answer requires recognizing group differences (yes) or whether groups should be treated identically (no). Crucially, fine-tuning for accuracy triggers harm drift: model-generated explanations become increasingly harmful as decision accuracy improves, whether by elaborating harmful content, introducing problematic assumptions, or failing to flag harms the baseline identified. To mitigate this, we introduce DART (Distill--Audit--Repair Training), which distills label-conditioned reasoning from a teacher, audits outputs for harm drift cases relative to baseline, and repairs problematic cases via severity-weighted fine-tuning. On eight benchmarks, DART improves Llama-3-8B-Instruct accuracy from 39.0% to 68.8%, with largest gains on equal-treatment prompts (11.3% -> 72.6%), while reducing harm drift cases by 72.6%. It also transfers to 280 open-ended real-world queries across medical, legal, policy, and educational domains, improving difference-appropriate responses from 39.8% to 77.5% while reducing refusals from 34.3% to 3.0%. Our results demonstrate that accuracy and safety need not conflict when explicit detection and repair mechanisms are in place.

Large language models (LLMs) are increasingly deployed not only to make decisions but to explain them. While AI decision fairness has been studied extensively, the fairness of AI explanations (whether LLMs justify decisions with equal quality, depth, tone, and linguistic sophistication across demographic groups) has received little attention. This paper introduces the Explanation Fairness Taxonomy (EFT), a framework comprising five formally defined, operationalizable dimensions: Verbosity Disparity, Sentiment Disparity, Epistemic Hedging Disparity, Decision-Linked Explanation Disparity, and Lexical Complexity Disparity. The taxonomy is instantiated in a controlled empirical study across 80 prompt templates, four consequential decision domains (hiring, medical triage, credit assessment, legal judgment), and five LLMs: GPT-4.1, Claude Sonnet, LLaMA 3.3 70B, GPT-OSS 120B, and Qwen3 32B. Two novel black-box metrics are introduced: the Hedging Density Score (HDS) and the Explanation Faithfulness Proxy (EFP), a heuristic indicator of decision-linked explanation variation. Across up to 400 prompt pairs, all eight EFT metrics show statistically significant disparities (Cohen's d ranging from small to large, all p_BH < 10^(-62)). Model choice is strongly associated with disparity magnitude: Qwen3 32B exhibits verbosity disparities 5.9x larger than LLaMA 3.3 70B. Two prompting-based mitigations show significant reductions in EFP disparity (78-95%) but no significant effect on stylistic dimensions, consistent with the hypothesis that stylistic explanation inequalities are encoded in pre-training distributions and are not resolvable through deployment-level instruction alone. A reproducible measurement framework is offered for explanation-level fairness auditing, with implications for AI regulation and deployment practice.

Fair allocation of scarce, indivisible resources is an important challenge in many societal problems. While there are several formal theories of fairness, no single definition can always be satisfied. As large language models (LLMs) are increasingly used to support decisions and act as agents, they raise new concerns about distributional justice: their judgments are not directly tied to any specific fairness framework and may violate key normative principles. In this work, we introduce a general method for evaluating fairness reasoning in LLMs. We study first-person fairness judgments across a broad set of models and compare them directly with human responses on matched scenarios and elicitation conditions. We find that LLMs tend to prefer stricter fairness constraints than humans, show more self-interested behavior, are sensitive to how information is framed, and are difficult to align with human judgments using fine-tuning with current datasets.