EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence
arXiv cs.AI · 23 Oct 2026, 03:57 AEST
EpiSEM-Bench is a new benchmark designed to evaluate the safety and evidence recovery of language model agents when critical information is missing. It highlights significant variations in how different models handle evidence omission and their ability to perform safely.
Key facts
- Tool-using language model agents rely on environmental facts to decide whether an action is permitted.
- We introduce ench{}, an executable benchmark for evaluating unsafe execution, safety boundaries, and evidence recovery in LLM agents under missing evidence.
- The benchmark reveals substantial variation across the ten models in omission robustness, evidence acquisition, and appropriate action.
Read aloud
EpiSEM-Bench is a new benchmark designed to evaluate the safety and evidence recovery of language model agents when critical information is missing. It highlights significant variations in how different models handle evidence omission and their ability to perform safely. Tool-using language model agents rely on environmental facts to decide whether an action is permitted. We introduce ench, an executable benchmark for evaluating unsafe execution, safety boundaries, and evidence recovery in L L M agents under missing evidence. The benchmark reveals substantial variation across the ten models in omission robustness, evidence acquisition, and appropriate action.
Sources in this cluster
- EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence
arXiv cs.AI · 23 Oct 2026, 03:57 AEST