Hacking or Just Failing? A More Capable LLM Monitor Better Separates Reward Hacking from Honest Failure
cs.AI, cs.LG
Submitted: 2026-10-30
Updated: 2026-10-30
Code: https://github.com/KaushikKC/weak-monitor-lab
Terminology
Sources
- From Confident Closing to Silent Failure: Characterizing False Success in LLM Agents
- RewardHackingAgents: Benchmarking Evaluation Integrity for LLM ML-Engineering Agents
- Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysis
- EvilGenie: A Reward Hacking Benchmark
- AI Control: Improving Safety Despite Intentional Subversion
- Reliable Weak-to-Strong Monitoring of LLM Agents
- Do Prompt-Elicited Trajectories Reflect Training-Time Reward Hacking? A Systematic Study on Monitoring Training-Time Reward Hacking in Code Generation
- Classifier Context Rot: Monitor Performance Degrades with Context Length
- TRACE: Trajectory Reasoning through Adaptive Cross-Step Evidence Aggregation for LLM Agents
- Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Use
- SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents
- BenchShield: Formal Model-Backed Instrumentation for Reward Integrity in LLM-Agent Evaluation Infrastructure
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents