CheatBench: Measuring Reward Gaming in AI Agents
On the map: Deep reinforcement learning methods → RL training of language and agent models
Abstract (arXiv)
Reinforcement learning has helped AI agents solve increasingly difficult tasks, but high rewards do not always reflect the work users intended. In recent incidents and controlled evaluations across the AI industry, agents trained to maximize reward have accessed unauthorized information, attempted to evade monitoring systems, and even breached sandbox protections to attack external systems. As agents become more capable, this behavior could pose increasingly serious risks. To measure this problem, we introduce CheatBench, a benchmark of cheating in AI agents across mathematical research, knowledge work, coding, visual tasks, and other domains. Its environments combine challenging assignments with opportunities to cheat, allowing researchers to study how agents pursue a goal when honest work is difficult. CheatBench supports comparisons across models and task categories, providing a testbed for measuring and reducing cheating as agents take on more consequential responsibilities. We publicly release CheatBench at https://cheatbench.ai
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods. Builds on 18 of its 34 references that are in the map. First posted on arXiv 28 Sep 2026.
Builds on
18 of 34 references.
References · 18
- 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
- 2022Scaling Laws for Reward Model Overoptimizationcited by 1,095
- 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
- 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
- 2022The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Modelscited by 328
- 2024Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Modelscited by 178
- 2025GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Taskscited by 158
- 2019Reward Tampering Problems and Solutions in Reinforcement Learning: A Causal Influence Diagram Perspectivecited by 108
- 2025Measuring Sycophancy of Language Models in Multi-turn Dialoguescited by 104
- 2017Reinforcement Learning with a Corrupted Reward Channelcited by 70
- 2025ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Casescited by 65
- 2020Hidden Incentives for Auto-Induced Distributional Shiftcited by 43
6 more
- 2026Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Usecited by 31
- 2026Agents' Last Examcited by 31
- 2025EvilGenie: A Reward Hacking Benchmarkcited by 28
- 2026Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scalecited by 10
- 2026BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Taskscited by 4
- 2026Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Valuescited by 3
Built on by
Citing papers per year: 2026: 1.
1 paper.
- 2026REA: From Correct LLM Judgments to Safe Decision Delivery12 Oct 2026
Similar, not cited
Close in wording, no citation either way
- 2026Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack
- 2026Reward Hacking as Equilibrium under Finite Evaluation
- 2026Do Coding Agents Deceive Us? Detecting and Preventing Cheating via Capped Evaluation with Randomized Tests
- 2026Reward Hacking in Language Model Agents: Revisiting AI Safety Gridworlds
- 2026Hack-Verifiable Terminal Bench: Evaluating Reward Hacking in Terminal Tasks
- 2026BenchShield: Formal Model-Backed Instrumentation for Reward Integrity in LLM-Agent Evaluation Infrastructure
- 2018Scalable agent alignment via reward modeling: a research direction
- 2026Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVR