AI sanctions screening study: The OFAC List Said Yes. Google AI Said No.

Abstract

A company-authored case study of source verification and outcome classification in sanctions-status research. A deliberately selected panel contained 64 companies removed from the OFAC SDN List and 32 listed controls. In 96 Google searches on 8 October 2026, 68 answers gave correct current status and 17 gave wrong status; abstention, historical-only, broader-scope and missing-summary outcomes are reported separately. The page also reports separate API experiments with distinct denominators. Archived answers, source records, review notes and offline verification materials support inspection. Useful for teaching measurement validity, provenance, uncertainty, changing source versions and limits of generalization. It is not peer reviewed and does not establish a population error rate, screening accuracy, ownership restrictions or transaction permission. This is a different study from our sanctions-list field-availability resource already in the catalog.