๐
๐
Old Age
DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training
April 18, 2026 ยท Grace Period ยท ๐ Findings of ACL 2026
Authors
Ziwen Pan, Zihan Liang, Jad Kabbara, Ali Emami
arXiv ID
2604.16845
Category
cs.CL: Computation & Language
Citations
0
Venue
Findings of ACL 2026
Abstract
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.
Community Contributions
Found the code? Know the venue? Think something is wrong? Let us know!
๐ Similar Papers
In the same crypt โ Computation & Language
๐
๐
Old Age
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
๐
๐
Old Age
XLNet: Generalized Autoregressive Pretraining for Language Understanding
๐ฎ
๐ฎ
The Ethereal
Effective Approaches to Attention-based Neural Machine Translation
๐
๐
Old Age
A large annotated corpus for learning natural language inference
๐
๐
Old Age