# DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training

arXiv: https://arxiv.org/abs/2604.16845 · HTML: https://mlsanity.com/paper/2604.16845/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 31 of its 68 references that are in the map.
- First posted on arXiv 18 Apr 2026.

## Builds on (most cited first, 30 of 31)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Self-Instruct: Aligning Language Models with Self-Generated Instructions](https://mlsanity.com/paper/2212.10560/) (2022)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model](https://mlsanity.com/paper/2405.04434/) (2024)
- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes](https://mlsanity.com/paper/2305.02301/) (2023)
- [Safe RLHF: Safe Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2310.12773/) (2023)
- [Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training](https://mlsanity.com/paper/2401.05566/) (2024)
- [Zephyr: Direct Distillation of LM Alignment](https://mlsanity.com/paper/2310.16944/) (2023)
- [Conflict-Averse Gradient Descent for Multi-task Learning](https://mlsanity.com/paper/2110.14048/) (2021)
- [ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection](https://mlsanity.com/paper/2203.09509/) (2022)
- [BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation](https://mlsanity.com/paper/2101.11718/) (2021)
- [Teaching Small Language Models to Reason](https://mlsanity.com/paper/2212.08410/) (2022)
- [HateCheck: Functional Tests for Hate Speech Detection Models](https://mlsanity.com/paper/2012.15606/) (2020)
- [Navigate through Enigmatic Labyrinth A Survey of Chain of Thought Reasoning: Advances, Frontiers and Future](https://mlsanity.com/paper/2309.15402/) (2023)
- ["I'm sorry to hear that": Finding New Biases in Language Models with a Holistic Descriptor Dataset](https://mlsanity.com/paper/2205.09209/) (2022)
- [Orca 2: Teaching Small Language Models How to Reason](https://mlsanity.com/paper/2311.11045/) (2023)
- [Co-Writing with Opinionated Language Models Affects Users' Views](https://mlsanity.com/paper/2302.00560/) (2023)
- [What Large Language Models Know and What People Think They Know](https://mlsanity.com/paper/2401.13835/) (2024)
- [SCOTT: Self-Consistent Chain-of-Thought Distillation](https://mlsanity.com/paper/2305.01879/) (2023)
- [Challenges in Automated Debiasing for Toxic Language Detection](https://mlsanity.com/paper/2102.00086/) (2021)
- [Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templates](https://mlsanity.com/paper/2402.18540/) (2024)
- [Probing LLMs for hate speech detection: strengths and vulnerabilities](https://mlsanity.com/paper/2310.12860/) (2023)
- [Investigating Subtler Biases in LLMs: Ageism, Beauty, Institutional, and Nationality Bias in Generative Models](https://mlsanity.com/paper/2309.08902/) (2023)
- [From Judgment to Interference: Early Stopping LLM Harmful Outputs via Streaming Content Monitoring](https://mlsanity.com/paper/2506.09996/) (2025)
- [Fairness through Difference Awareness: Measuring Desired Group Discrimination in LLMs](https://mlsanity.com/paper/2502.01926/) (2025)
- [Text-Diffusion Red-Teaming of Large Language Models: Unveiling Harmful Behaviors with Proximity Constraints](https://mlsanity.com/paper/2501.08246/) (2025)

## Abstract (arXiv)

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.