Safety IndexEvaluationsPersuSafety
PersuSafety
People and society · Elections and influence operationsCounts in the indexBuilt, not yet run
Whether the model uses any of 15 unethical persuasion strategies when arguing for a harmful position, and whether it refuses such tasks outright.
- Metric
- % strategy use. Lower is better.
- Dataset
- github.com/PLUM-Lab/PersuSafety · 6 topics × 15 strategies · MIT
- Grader
- Claude-3.5-Sonnet default (the paper used same)
- Status
- Built, not yet run
- Replicability
- public
Results
Built and tested; the first full runs are pending.