Better Call Reward: Reward Hacking as Strategic Abstention in Legal Reasoning Models

arXiv:2610.06439 · cs.LG, cs.AI, cs.CL, cs.CY · Submitted 2026-10-05 · Read on arXiv

cs.LG, cs.AI, cs.CL, cs.CY

Submitted: 2026-10-05

Updated: 2026-10-05

Code: https://github.com/SubramanyamSahoo/Better-Call-Reward

Terminology

• Abstention — 1× in this paper · explained in 1 episode
• Reward Hacking — 1× in this paper · explained in 8 episodes

Sources

• Scaling Laws for Reward Model Overoptimization
• LegalBench: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Models
• Legal Alignment for Safe and Ethical AI
• Training language models to follow instructions with human feedback
• The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models
• Pessimism's Paradox: Conservative Offline Training Amplifies Reward Hacking During Online Adaptation in Reasoning Models
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• Defining and Characterizing Reward Hacking
• Qwen3 Technical Report

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks