HoneyBench - A general benchmark for reward hacking in frontier models
Key takeaway
We've just released the first version of HoneyBench.
Why it's on PDOOM
PDOOM selected this story for its alignment & control signals: Alignment, Reward Hacking, Benchmark.
AlignmentReward HackingBenchmarkFrom LessWrong AI
We've just released the first version of HoneyBench. It consists of nine 'honeypot' challenges that each elicit unique and antisocial examples of reward hacking from some or all of major labs’ top public releases, including Opus 5.5, Fable 5.1, GPT-6 Astra, Gemini 3.8 Flash, Grok 4.7, and DeepSeek V4 Pro.All modern large language models engage in some degree of specification gaming, both during and outside training. High-quality alignment evaluations, in combination with other techniques such as
By Dean Valentine