Governing AI Through Public Procurement

• In Event: Collaborative Governance in the Digital Era: Accountability, Deliberation and Decentralization

Abstract

Public agencies increasingly adopt artificial intelligence through contracts, vendor agreements, solicitations, and acquisition rules rather than through standalone legislation. As a result, procurement is becoming one of the most important but understudied sites of AI governance in the public sector. Decisions about auditability, transparency, bias mitigation, data rights, safety restrictions, and vendor responsibility are often embedded in procurement documents rather than in publicly visible policy debates. This raises an important public administration question: can procurement function as a meaningful governance mechanism for AI, or does it simply relocate major accountability choices into opaque administrative processes? This paper asks: Under what conditions does public procurement strengthen or weaken accountability in government AI adoption, and what design features of procurement systems matter most for transparency, oversight, and public protection? The study draws on OMB Memorandum M-25-22 on AI acquisition guidance; the March 2026 GSA proposed clause “Basic Safeguarding of Artificial Intelligence Systems”; award and procurement data from USASpending.gov; solicitation and contract documents from SAM.gov; and selected municipal procurement materials and AI-related contract documents from local governments with active digital governance initiatives, including Seattle and New York City, where publicly available. Procurement documents were identified through purposive sampling of federal and municipal cases that explicitly reference AI, automated systems, model oversight, auditability, or vendor accountability, with cases selected to capture variation in procurement structure and governance language. Using a qualitative institutional and document-based analysis, the paper examines how procurement documents structure accountability for AI systems. It compares procurement language across federal and selected local cases, focusing on transparency requirements, audit access, model limitations, indemnification, safety restrictions, and allocation of responsibility between agencies and vendors. The analysis is informed by theories of governance through contracts and public accountability. Preliminary analysis suggests that procurement is not merely an administrative tool for acquiring AI systems; it is a substantive governance mechanism that determines how risks are defined, who can inspect systems, and how responsibility is distributed. Early evidence indicates that some procurement frameworks strengthen accountability by requiring disclosure, audit access, and documentation, while others leave critical oversight questions vague or defer heavily to vendor terms. In practice, procurement may serve as one of the last meaningful guardrails when broader regulatory frameworks are weak or fragmented. The paper argues that procurement should be treated as a central site of AI governance in public administration. This shifts attention away from abstract debates about AI ethics and toward the concrete institutional mechanisms through which governments shape accountability in implementation. For APPAM and policy audiences, the implication is that governing AI effectively requires closer scrutiny of acquisition systems, not just downstream outcomes or high-level regulatory statements.

Author

• Albert Nii Noi Okwei, Virginia Commonwealth University
  Presenting Author