# Dissecting Bias in LLMs: A Mechanistic Interpretability Perspective

arXiv: https://arxiv.org/abs/2506.05166 · HTML: https://mlsanity.com/paper/2506.05166/

## Key facts

- Cited by 15 papers in the map (data through 14 Oct 2026).
- Placed in Probing deception, bias and beliefs, one of 13 directions in Mechanistic interpretability.
- Builds on 25 of its 38 references that are in the map.
- First posted on arXiv 5 Jun 2025.

## Placed in

- Mechanistic interpretability / [Probing deception, bias and beliefs](https://mlsanity.com/mechinterp/probing-deception-bias-and-beliefs/)

## Builds on (most cited first, 25 of 25)

- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings](https://mlsanity.com/paper/1607.06520/) (2016)
- [Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small](https://mlsanity.com/paper/2211.00593/) (2022)
- [Neural Network Acceptability Judgments](https://mlsanity.com/paper/1805.12471/) (2018)
- [Progress measures for grokking via mechanistic interpretability](https://mlsanity.com/paper/2301.05217/) (2023)
- [Towards Automated Circuit Discovery for Mechanistic Interpretability](https://mlsanity.com/paper/2304.14997/) (2023)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [Causal Abstractions of Neural Networks](https://mlsanity.com/paper/2106.02997/) (2021)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [Attribution Patching Outperforms Automated Circuit Discovery](https://mlsanity.com/paper/2310.10348/) (2023)
- [Counterfactual Data Augmentation for Mitigating Gender Stereotypes in Languages with Rich Morphology](https://mlsanity.com/paper/1906.04571/) (2019)
- [Localizing Model Behavior with Path Patching](https://mlsanity.com/paper/2304.05969/) (2023)
- [Gender bias and stereotypes in Large Language Models](https://mlsanity.com/paper/2308.14921/) (2023)
- [Interpretability at Scale: Identifying Causal Mechanisms in Alpaca](https://mlsanity.com/paper/2305.08809/) (2023)
- [Self-Debiasing Large Language Models: Zero-Shot Recognition and Reduction of Stereotypes](https://mlsanity.com/paper/2402.01981/) (2024)
- [Nationality Bias in Text Generation](https://mlsanity.com/paper/2302.02463/) (2023)
- [Investigating Subtler Biases in LLMs: Ageism, Beauty, Institutional, and Nationality Bias in Generative Models](https://mlsanity.com/paper/2309.08902/) (2023)
- [Backward Lens: Projecting Language Model Gradients into the Vocabulary Space](https://mlsanity.com/paper/2402.12865/) (2024)
- [Linear Representations of Political Perspective Emerge in Large Language Models](https://mlsanity.com/paper/2503.02080/) (2025)
- [Gender Biases in Automatic Evaluation Metrics for Image Captioning](https://mlsanity.com/paper/2305.14711/) (2023)
- [Identifying and Adapting Transformer-Components Responsible for Gender Bias in an English Language Model](https://mlsanity.com/paper/2310.12611/) (2023)
- [ChatGPT Based Data Augmentation for Improved Parameter-Efficient Debiasing of LLMs](https://mlsanity.com/paper/2402.11764/) (2024)
- [BiasEdit: Debiasing Stereotyped Language Models via Model Editing](https://mlsanity.com/paper/2503.08588/) (2025)
- [Quantifying and Reducing Stereotypes in Word Embeddings](https://mlsanity.com/paper/1606.06121/) (2016)
- [The Impact of Debiasing on the Performance of Language Models in Downstream Tasks is Underestimated](https://mlsanity.com/paper/2309.09092/) (2023)

## Built on by (most cited first, 15 of 15)

- [Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models](https://mlsanity.com/paper/2601.14004/) (2026)
- [Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability](https://mlsanity.com/paper/2510.12229/) (2025)
- [Explainability of Large Language Models: Opportunities and Challenges toward Generating Trustworthy Explanations](https://mlsanity.com/paper/2510.17256/) (2025)
- [The Homogenization Problem in LLMs: Towards Meaningful Diversity in AI Safety](https://mlsanity.com/paper/2601.06116/) (2026)
- [Attention, May I Have Your Decision? Localizing Generative Choices in Diffusion Models](https://mlsanity.com/paper/2604.06052/) (2026)
- [BiasGym: A Simple and Generalizable Framework for Analyzing and Removing Biases through Injection](https://mlsanity.com/paper/2508.08855/) (2025)
- [The Psychology of Falsehood: A Human-Centric Survey of Misinformation Detection](https://mlsanity.com/paper/2509.15896/) (2025)
- [Measuring Mechanistic Independence: Can Bias Be Removed Without Erasing Demographics?](https://mlsanity.com/paper/2512.20796/) (2025)
- [Towards general embodied intelligence: integrating large language models, knowledge bases, and reasoning capabilities to build the next generation of AI agents](https://mlsanity.com/paper/2608.19794/) (2026)
- [White-Box Sensitivity Auditing with Steering Vectors](https://mlsanity.com/paper/2601.16398/) (2026)
- [UGID: Unified Graph Isomorphism for Debiasing Large Language Models](https://mlsanity.com/paper/2603.19144/) (2026)
- [GKnow: Measuring the Entanglement of Gender Bias and Factual Gender](https://mlsanity.com/paper/2605.12299/) (2026)
- [A Mechanistic Understanding of Pronoun Fidelity in LLMs](https://mlsanity.com/paper/2606.16407/) (2026)
- [S^3martCirc: Self-supervised Smart Circuit Discovery](https://mlsanity.com/paper/2609.00755/) (2026)
- [Thin Evidence, Thick Priors: How Language Models Substitute Identity for Missing Financial Facts](https://mlsanity.com/paper/2610.07798/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) are known to exhibit social, demographic, and gender biases, often as a consequence of the data on which they are trained. In this work, we adopt a mechanistic interpretability approach to analyze how such biases are structurally represented within models such as GPT-2 and Llama2. Focusing on demographic and gender biases, we explore different metrics to identify the internal edges responsible for biased behavior. We then assess the stability, localization, and generalizability of these components across dataset and linguistic variations. Through systematic ablations, we demonstrate that bias-related computations are highly localized, often concentrated in a small subset of layers. Moreover, the identified components change across fine-tuning settings, including those unrelated to bias. Finally, we show that removing these components not only reduces biased outputs but also affects other NLP tasks, such as named entity recognition and linguistic acceptability judgment because of the sharing of important components with these tasks.