# Misaligned by Reward: Socially Undesirable Preferences in LLMs

arXiv: https://arxiv.org/abs/2605.05003 · HTML: https://mlsanity.com/paper/2605.05003/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 29 of its 41 references that are in the map.
- First posted on arXiv 6 May 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Human values and AI feedback alignment](https://mlsanity.com/alignment/human-values-and-ai-feedback-alignment/)

## Builds on (most cited first, 29 of 29)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Qwen Technical Report](https://mlsanity.com/paper/2309.16609/) (2023)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [StereoSet: Measuring stereotypical bias in pretrained language models](https://mlsanity.com/paper/2004.09456/) (2020)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Models](https://mlsanity.com/paper/2010.00133/) (2020)
- [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)
- [RewardBench: Evaluating Reward Models for Language Modeling](https://mlsanity.com/paper/2403.13787/) (2024)
- [Gender Bias in Coreference Resolution](https://mlsanity.com/paper/1804.09301/) (2018)
- [ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection](https://mlsanity.com/paper/2203.09509/) (2022)
- [Understanding Dataset Difficulty with V-Usable Information](https://mlsanity.com/paper/2110.08420/) (2021)
- [PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference](https://mlsanity.com/paper/2406.15513/) (2024)
- [A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity](https://mlsanity.com/paper/2401.01967/) (2024)
- [Moral Stories: Situated Reasoning about Norms, Intents, Actions, and their Consequences](https://mlsanity.com/paper/2012.15738/) (2020)
- [MaxMin-RLHF: Alignment with Diverse Human Preferences](https://mlsanity.com/paper/2402.08925/) (2024)
- [M-RewardBench: Evaluating Reward Models in Multilingual Settings](https://mlsanity.com/paper/2410.15522/) (2024)
- [From Instructions to Intrinsic Human Values -- A Survey of Alignment Goals for Big Models](https://mlsanity.com/paper/2308.12014/) (2023)
- [Bias of AI-Generated Content: An Examination of News Produced by Large Language Models](https://mlsanity.com/paper/2309.09825/) (2023)
- [Transforming and Combining Rewards for Aligning Large Language Models](https://mlsanity.com/paper/2402.00742/) (2024)
- [On the Robustness of Reward Models for Language Model Alignment](https://mlsanity.com/paper/2505.07271/) (2025)
- [Detecting Prefix Bias in LLM-based Reward Models](https://mlsanity.com/paper/2505.13487/) (2025)
- [Towards Reward Fairness in RLHF: From a Resource Allocation Perspective](https://mlsanity.com/paper/2505.23349/) (2025)
- [Guiding LLM Decision-Making with Fairness Reward Models](https://mlsanity.com/paper/2507.11344/) (2025)
- [Towards Large Language Models that Benefit for All: Benchmarking Group Fairness in Reward Models](https://mlsanity.com/paper/2503.07806/) (2025)

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

- [Which Objectives Need a Dial? Predicting Objective Conflict and Covering Trade-offs in Steerable Pluralistic Alignment](https://mlsanity.com/paper/2609.26929/) (2026)

## Abstract (arXiv)

Reward models are a key component of large language model alignment, serving as proxies for human preferences during training. However, existing evaluations focus primarily on broad instruction-following benchmarks, providing limited insight into whether these models capture socially desirable preferences. As a result, important failures in social alignment can remain hidden. We extend reward-model benchmarking to four socially consequential domains: bias, safety, morality, and ethical reasoning. We introduce a framework that converts social evaluation datasets into pairwise preference data, leveraging gold labels where available and directional bias indicators otherwise. This enables us to test whether reward models prefer socially undesirable responses, and whether their preferences produce systematically biased distributions over selected outputs. Across five publicly available reward models and two instruction-tuned models used as reward proxies, we find substantial variation across domains, with no single model performing best overall. The models fall well short of strong social intelligence: they often prefer socially undesirable options, and their preferences produce systematically biased distributions. Moreover, stronger bias avoidance can reduce sensitivity to context, revealing a key alignment trade-off between avoiding biased outcomes and preserving contextual faithfulness. These findings show that standard reward benchmarks are insufficient for assessing social alignment and highlight the need for evaluations that directly measure the social preferences encoded in reward models.