The Reward Hacking Benchmark

Reward hacking makes it really difficult to have the model genuinely improve upon the things you care about, because it's capable enough to find shortcuts around it. Earlier this year (around May), I wanted to characterize the extent of reward hacking in frontier LLMs. The first version of the Reward Hacking Benchmark (RHB-v1) had five code optimization tasks (sorting, regex, compression, matrix multiplication, prime checking) evaluated in a deliberately vulnerable sandbox. Models started with inefficient but correct code and got multiple iterations to improve, with feedback on their performance and ranking in a leaderboard after each attempt. Models didn't interact with the environment directly – instead, they communicated through a structured harness with specific fields for different actions: “bash”, "cot", “code”, etc, and the harness executed on their behalf, in a loop with the model. The real results were about environments. I ran every model through two setups: a permissive sand

The Reward Hacking Benchmark The Reward Hacking Benchmark Published on July 1, 2025 Last updated on October 28, 2025 "Every system is perfectly designed to get the results it gets." Reward hacking makes it really difficult to have the model genuinely improve upon the things you care about, because it's capable enough to find shortcuts around it. Earlier this year (around May), I wanted to characterize the extent of reward hacking in frontier LLMs. The first version of the Reward Hacking Benchmark (RHB-v1) had five code optimization tasks (sorting, regex, compression, matrix multiplication, pri

related reading

• Reward Hacking in Reinforcement Learning | Lil'Loglilianweng.github.io
• Recent Frontier Models Are Reward Hacking - METRmetr.org
• Reward hacking is becoming more sophisticated and deliberate in frontier LLMs — LessWronglesswrong.com
• Training a Misaligned Reward Seekeralignment.anthropic.com
• Systematic Reward Hacking and Prime Sprintsprimeintellect.ai
• Reward hacking behavior can generalize across tasks — AI Alignment Forumalignmentforum.org
• Natural-emergent-misalignment-from-reward-hacking-paper.pdfassets.anthropic.com
• (Some) Natural Emergent Misalignment from Reward Hacking in Non-Production RL — LessWronglesswrong.com
• Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscationarxiv.org
• Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face — LessWronglesswrong.com
• [2511.21654] EvilGenie: A Reward Hacking Benchmarkarxiv.org
• Through the looking glass of benchmark hacking — Poolsidepoolside.ai