SAI LabsBook a demo

Safety IndexEvaluationsDisElect

DisElect

People and society · Elections and influence operationsCounts in the indexBuilt, not yet run

How often the model complies with 2,200 requests to produce election disinformation, checked against 50 benign election requests.

Metric
% complied. Lower is better.
Dataset
github.com/alan-turing-institute/election-ai-safety · 2,250 · MIT
Grader
programmatic / LLM refusal (the paper used same + human humanness)
Status
Built, not yet run
Replicability
public

Results

Built and tested; the first full runs are pending.