# Fair outputs, Biased Internals: Causal Potency and Asymmetry of Latent Bias in LLMs for High-Stakes Decisions

arXiv: https://arxiv.org/abs/2605.15217 · HTML: https://mlsanity.com/paper/2605.15217/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 24 of its 36 references that are in the map.
- First posted on arXiv 12 May 2026.

## Builds on (most cited first, 24 of 24)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings](https://mlsanity.com/paper/1607.06520/) (2016)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models](https://mlsanity.com/paper/2305.18189/) (2023)
- [Empowering Many, Biasing a Few: Generalist Credit Scoring through Large Language Models](https://mlsanity.com/paper/2310.00566/) (2023)
- [Benchmarking Adversarial Robustness to Bias Elicitation in Large Language Models: Scalable Automated Assessment with LLM-as-a-Judge](https://mlsanity.com/paper/2504.07887/) (2025)
- [LLMs are Vulnerable to Malicious Prompts Disguised as Scientific Language](https://mlsanity.com/paper/2501.14073/) (2025)
- [Divine LLaMAs: Bias, Stereotypes, Stigmatization, and Emotion Representation of Religion in Large Language Models](https://mlsanity.com/paper/2407.06908/) (2024)
- [Silenced Biases: The Dark Side LLMs Learned to Refuse](https://mlsanity.com/paper/2511.03369/) (2025)
- [What's Not Said Still Hurts: A Description-Based Evaluation Framework for Measuring Social Bias in LLMs](https://mlsanity.com/paper/2502.19749/) (2025)
- [Forecasting Credit Ratings: A Case Study where Traditional Methods Outperform Generative LLMs](https://mlsanity.com/paper/2407.17624/) (2024)
- [More Women, Same Stereotypes: Unpacking the Gender Bias Paradox in Large Language Models](https://mlsanity.com/paper/2503.15904/) (2025)
- [Interpretable LLMs for Credit Risk: A Systematic Review and Taxonomy](https://mlsanity.com/paper/2506.04290/) (2025)
- [On Optimal Steering to Achieve Exact Fairness](https://mlsanity.com/paper/2509.15759/) (2025)
- [Interpreting LLMs as Credit Risk Classifiers: Do Their Feature Explanations Align with Classical ML?](https://mlsanity.com/paper/2510.25701/) (2025)
- [Alignment Reduces Expressed but Not Encoded Gender Bias: A Unified Framework and Study](https://mlsanity.com/paper/2603.24125/) (2026)
- [A Large Language Model for Corporate Credit Scoring](https://mlsanity.com/paper/2511.02593/) (2025)
- [LendNova: Towards Automated Credit Risk Assessment with Language Models](https://mlsanity.com/paper/2601.02573/) (2026)

## Built on by (most cited first, 1 of 1)

- [Locating and Controlling Implicit Personalization in Large Language Models](https://mlsanity.com/paper/2608.11735/) (2026)

## Abstract (arXiv)

Instruction-tuned language models exhibit behavioural fairness in high-stakes decisions while retaining biased associations in their internal representations. However, whether these suppressed representations can affect model outputs - and whether such causal potency is symmetric across demographic groups - remains unknown. We investigate the use of open-weight models for mortgage underwriting using matched applications that differ only in racially-associated names and reveal a critical disconnect: models show no output-level bias, yet retain and amplify demographic representations across model layers. Through activation steering and novel cross-layer interventions, we demonstrate that this suppressed information is decision-relevant: when reinjected at critical layers, it produces near-complete decision reversals. Critically, this latent bias is asymmetric - steering interventions affect decisions in one demographic direction, while producing minimal effects in reverse - and susceptible to adversarial prompt engineering and parameter-efficient fine-tuning. These findings demonstrate that behavioural audits focused on outputs are insufficient: fair outputs can mask exploitable internal biases. They also motivate dual-layer testing frameworks combining output evaluation with representational analysis for AI governance in high-stakes decisions.