Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysis

TL;DR

This paper introduces TRACE, a benchmark for detecting reward hacking in code environments, revealing that contrastive analysis improves detection and highlighting challenges in identifying semantically contextualized hacks.

Contribution

The paper presents a new taxonomy of reward exploits, a synthetic benchmark TRACE, and demonstrates the effectiveness of contrastive analysis for reward hack detection in code environments.

Findings

Contrastive analysis improves reward hack detection rates.

GPT-5.2 achieves up to 63% detection accuracy.

Semantic context increases difficulty in detecting reward hacks.

Abstract

Recent advances in reinforcement learning for code generation have made robust environments essential to prevent reward hacking. As LLMs increasingly serve as evaluators in code-based RL, their ability to detect reward hacking remains understudied. In this paper, we propose a novel taxonomy of reward exploits spanning across 54 categories and introduce TRACE (Testing Reward Anomalies in Code Environments), a synthetically curated and human-verified benchmark containing 517 testing trajectories. Unlike prior work that evaluates reward hack detection in isolated classification scenarios, we contrast these evaluations with a more realistic, contrastive anomaly detection setup on TRACE. Our experiments reveal that models capture reward hacks more effectively in contrastive settings than in isolated classification settings, with GPT-5.2 with highest reasoning mode achieving the best detection rate at 63%, up from 45% in isolated settings on TRACE. Building on this insight, we demonstrate that state-of-the-art models struggle significantly more with semantically contextualized reward hacks compared to syntactically contextualized ones. We further conduct qualitative analyses of model behaviors, as well as ablation studies showing that the ratio of benign to hacked trajectories and analysis cluster sizes substantially impact detection performance. We release the benchmark and evaluation harness to enable the community to expand TRACE and evaluate their models.