CATCH Benchmark
Explore score history ↗During reinforcement learning with verifiable rewards (RLVR), large language models (LLMs) can exploit loopholes in their environments to obtain high rewards without improving the intended capabilities, i.e., reward hacking.
- Released
- 2026-09-30
- Institutions
- Tsinghua University · Beihang University · Southern University of Science and Technology
- Evaluates
- Cybersecurity, Language & Knowledge, cs.CL
- Openness
- unknown
- Importer
- Claire Radar
- Review status
- rule-auto-admitted
CATCH paper and code
No reported scores are on record for this benchmark yet.