[2603.03291] One Bias After Another: Mechanistic Reward Shaping and Persistent Biases in Language Reward Models

Computer Science > Computation and Language arXiv:2603.03291 (cs) [Submitted on 6 Feb 2026] Title:One Bias After Another: Mechanistic Reward Shaping and Persistent Biases in Language Reward Models Authors:Daniel Fein, Max Lamparth, Violet Xiang, Mykel J. Kochenderfer, Nick Haber View a PDF of the paper titled One Bias After Another: Mechanistic Reward Shaping and Persistent Biases in Language Reward Models, by Daniel Fein and 4 other authors View PDF HTML (experimental) Abstract:Reward Models (RMs) are crucial for online alignment of language models (LMs) with human preferences. However, RM-based preference-tuning is vulnerable to reward hacking, whereby LM policies learn undesirable behaviors from flawed RMs. By systematically measuring biases in five high-quality RMs, including the state-of-the-art, we find that issues persist despite prior work with respect to length, sycophancy, and overconfidence. We also discover new issues related to bias toward model-specific styles and answer-order. We categorize RM failures by complexity and propose a simple post-hoc intervention to mitigate low-complexity biases that arise from spurious correlations. Our proposed mechanistic reward shaping reduces targeted biases without degrading reward quality and while using minimal labeled data. The method is extensible to new biases, model-internal, and generalizes out-of-distribution. Comments: Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) Cite as: arXiv:2603....