# Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels

arXiv: https://arxiv.org/abs/2605.15208 · HTML: https://mlsanity.com/paper/2605.15208/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Builds on 17 of its 28 references that are in the map.
- First posted on arXiv 2 May 2026.

## Builds on (most cited first, 17 of 17)

- [A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions](https://mlsanity.com/paper/2311.05232/) (2023)
- [BBQ: A Hand-Built Bias Benchmark for Question Answering](https://mlsanity.com/paper/2110.08193/) (2021)
- [Bias and Fairness in Large Language Models: A Survey](https://mlsanity.com/paper/2309.00770/) (2023)
- [Large Language Models: A Survey](https://mlsanity.com/paper/2402.06196/) (2024)
- [What Do Compressed Deep Neural Networks Forget?](https://mlsanity.com/paper/1911.05248/) (2019)
- [Understanding and Overcoming the Challenges of Efficient Transformer Quantization](https://mlsanity.com/paper/2109.12948/) (2021)
- [The Dawn After the Dark: An Empirical Study on Factuality Hallucination in Large Language Models](https://mlsanity.com/paper/2401.03205/) (2024)
- [How Does Quantization Affect Multilingual LLMs?](https://mlsanity.com/paper/2407.03211/) (2024)
- [A Survey on Large Language Model Benchmarks](https://mlsanity.com/paper/2508.15361/) (2025)
- [Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression](https://mlsanity.com/paper/2407.04965/) (2024)
- [Accuracy is Not All You Need](https://mlsanity.com/paper/2407.09141/) (2024)
- [A Survey on Out-of-Distribution Evaluation of Neural NLP Models](https://mlsanity.com/paper/2306.15261/) (2023)
- [Understanding the Effect of Model Compression on Social Bias in Large Language Models](https://mlsanity.com/paper/2312.05662/) (2023)
- [Investigating the Impact of Quantization Methods on the Safety and Reliability of Large Language Models](https://mlsanity.com/paper/2502.15799/) (2025)
- [Robust Lottery Tickets for Pre-trained Language Models](https://mlsanity.com/paper/2211.03013/) (2022)
- [LLMCBench: Benchmarking Large Language Model Compression for Efficient Deployment](https://mlsanity.com/paper/2410.21352/) (2024)
- [Investigating Social Bias Changes in Quantized Language Models](https://mlsanity.com/paper/2602.06181/) (2026)

## Built on by (most cited first, 6 of 6)

- [Flat Score, Amplified Failures: How the Error Budget Masks Damage in Quantized LLM Agents](https://mlsanity.com/paper/2607.27275/) (2026)
- [QuantiBias: Benchmarking Quantization-Induced Bias in LLMs](https://mlsanity.com/paper/2607.21063/) (2026)
- [The Asymmetric Harms of LLM Compression](https://mlsanity.com/paper/2608.19670/) (2026)
- [Stress-Testing Efficient Responsible-AI Evaluation: When Compute Savings Change Benchmark Conclusions](https://mlsanity.com/paper/2608.31108/) (2026)
- [The Asymmetric Effects of Knowledge Distillation on Bias in Small Language Models](https://mlsanity.com/paper/2607.28639/) (2026)
- [Off-Target Effects of Response-Style Alignment in a Korean 27B Language Model](https://mlsanity.com/paper/2609.11291/) (2026)

## Abstract (arXiv)

Large Language Models are routinely compressed via post-training quantization to reduce inference costs and memory footprint for cloud and edge deployment, yet the impact of this compression on model quality remains poorly understood. Existing studies typically compare only two conditions (full-precision vs. a single quantized variant), rely on aggregate bias metrics, and evaluate a single model family, making it impossible to distinguish gradual degradation from threshold-dependent safety failures. We conduct a controlled empirical study of three instruction-tuned models (Qwen2.5-7B, Mistral-7B, Phi-3.5-mini) at five precision levels (BF16 through 3-bit) on 12,148 BBQ bias benchmark items across 5 random seeds, totaling 911,100 inference records. Our results reveal that 3-bit quantization causes 6-21% of previously unbiased items to develop new stereotypical behaviors, following a clear dose-response pattern confirmed via logistic regression, while models' willingness to select "unknown" answers declines by 17.4%. Crucially, these item-level changes are invisible to standard quality metrics: perplexity increases by less than 0.5% at 8-bit and under 3% at 4-bit across all three models, yet 2.5-5.6% of items already develop new biases at 4-bit. These findings demonstrate that aggregate evaluation metrics systematically miss fairness-critical degradation, underscoring the need for quality-aware compression protocols that explicitly test for bias emergence before deployment.