Auditing Generative AI Systems Procured by Government Agencies

Abstract

Background: Policy attention to generative AI has increased, yet there is limited empirical clarity on how institutional arrangements influence auditing and implementation quality.

Objective: The study evaluated institutional variation in approaches to generative AI, with particular attention to accountability, administrative capacity, and auditing.

Methods: The documentary sample comprised 738 statutes, regulations, policies, contracts, guidance documents, administrative records, or public disclosures. Two-stage coding evaluated clarity, transparency, procedural safeguards, oversight, and implementation requirements; disagreements were independently reconciled.

Results: The document review found formal generative AI requirements in 61% of sources, but detailed implementation or oversight provisions in just 41%. In 23% of cases, transparency or review mechanisms were missing or incomplete. Instruments with broader operational detail were associated with clearer responsibility and fewer implementation gaps.

Conclusion: For generative AI, the main practical lesson is that formal requirements do not by themselves produce consistent implementation. This means that procurement reform should prioritize traceable processes, coordinated oversight, and implementation capacity.

Study design: procurement audit study.

Publication history: Originally published in print on 15 June 2025, Volume 5, Article 2. Online abstract record added on 16 September 2026. Full-text PDF will be added when the digital galley is ready.