Space & aviationResearch findings
EpiSEM-Bench: Framework for Safety Evaluation and Evidence Recovery under Missing Evidence
Original published 2026.10.23 02:57 KSTAdded to Blackdeck 2026.10.26 14:53 KST
What has changed?
Introduces an executable benchmark with 300 matched unsafe/safe task pairs across six domains.
Key points
- Tool-using LLM agents rely on environmental facts to determine action permissibility.
- Omission of safety-relevant facts can lead to prohibited actions despite benign user requests.
- EpiSEM-Bench evaluates unsafe execution and evidence recovery under missing evidence.
Connect it to your work
Measures unsafe execution under critical evidence omission and evaluates recovery behavior in response to missing facts.
What remains unproven
This research note is based on an abstract; check the original for test conditions and application scope.
Originals used for this note
Links open the original publisher in a new tab.