Benchmark RadarIntegrityBench
IntegrityBench evaluates language models on research integrity tasks, including misconduct classification, ethical action reasoning, and artifact-grounded decision making, across 36 paired tasks with a 5-level pressure protocol spanning multiple domains and research stages.
- Released
- 2026-06-03
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
As language models are used as co-scientists, measuring their integrity under pressure is critical. This benchmark could inform deployment decisions and identify risks of facilitating misconduct or eroding trust, but the evaluation method and reproducibility are not yet specified.
Motivation
Language models are increasingly deployed as co-scientists, yet their ability to uphold research integrity under institutional pressure remains unmeasured.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.