# Who Do Language Models Think Is Competent? A Mechanistic Analysis of Occupational Bias

arXiv: https://arxiv.org/abs/2608.20347 · HTML: https://mlsanity.com/paper/2608.20347/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 14 of its 59 references that are in the map.
- First posted on arXiv 15 Jun 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 14 of 14)

- [Persona Vectors: Monitoring and Controlling Character Traits in Language Models](https://mlsanity.com/paper/2507.21509/) (2025)
- [Function Vectors in Large Language Models](https://mlsanity.com/paper/2310.15213/) (2023)
- [ReFT: Representation Finetuning for Language Models](https://mlsanity.com/paper/2404.03592/) (2024)
- [The Remarkable Robustness of LLMs: Stages of Inference?](https://mlsanity.com/paper/2406.19384/) (2024)
- [Evaluating and Mitigating Discrimination in Language Model Decisions](https://mlsanity.com/paper/2312.03689/) (2023)
- [Racial Disparity in Natural Language Processing: A Case Study of Social Media African-American English](https://mlsanity.com/paper/1707.00061/) (2017)
- [Detecting High-Stakes Interactions with Activation Probes](https://mlsanity.com/paper/2506.10805/) (2025)
- [Circuit Breaking: Removing Model Behaviors with Targeted Ablation](https://mlsanity.com/paper/2309.05973/) (2023)
- [Reading Between the Prompts: How Stereotypes Shape LLM's Implicit Personalization](https://mlsanity.com/paper/2505.16467/) (2025)
- [Robustly Improving LLM Fairness in Realistic Settings via Interpretability](https://mlsanity.com/paper/2506.10922/) (2025)
- [Investigating task-specific prompts and sparse autoencoders for activation monitoring](https://mlsanity.com/paper/2504.20271/) (2025)
- [Detecting (Un)answerability in Large Language Models with Linear Directions](https://mlsanity.com/paper/2509.22449/) (2025)
- [CRISP: Persistent Concept Unlearning via Sparse Autoencoders](https://mlsanity.com/paper/2508.13650/) (2025)
- [Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretability](https://mlsanity.com/paper/2602.10067/) (2026)

## Abstract (arXiv)

Language models (LMs) often pass behavioral bias evaluations, but it remains unclear whether they no longer represent the underlying associations that give rise to biases, or have merely learned not to express them. In this study, we show that representational biases are often detectable, even when behavioral biases are not visible. We introduce a causal framework that decomposes occupational bias into two measurement points: a model's internal representation of a user's competence, and its observable outputs. We derive steering vectors for representations of user expertise, and verify that they causally mediate model behavior in both a question-answering task and a hiring task. Applying this framework to several open-weight models, we find that demographic attributes, such as gender, race, and socioeconomic status, influence a model's representation of user expertise, even in cases where behavioral metrics detect no disparity between demographics. We show that these model representations can influence downstream behavior under intervention, suggesting failure modes that behavioral metrics alone may not detect.