CATCH: A Controllable Analysis Testbed for Reward Hacking in Coding RL
cs.CL, cs.LG
Submitted: 2026-09-30
Updated: 2026-10-02
Code: https://github.com/THUAIS-Lab/CATCH
Terminology
Sources
- Gradient Regularization Mitigates Reward Hacking in Reinforcement Learning from Human Feedback and Verifiable Rewards
- Concrete Problems in AI Safety
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- EvilGenie: A Reward Hacking Benchmark
- Skywork Open Reasoner 1 Technical Report
- Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVR
- Natural Emergent Misalignment from Reward Hacking in Production RL
- Defining and Characterizing Reward Hacking
- Detecting and Suppressing Reward Hacking with Gradient Fingerprints
- Qwen3 Technical Report
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering