State Variation in Public-Sector AI Governance

• In Event: Artificial Intelligence and Governance

Abstract

Artificial intelligence governance in the United States is increasingly shaped by state-level divergence. As federal and state authorities are actively contested, states are moving in different directions on public-sector AI regulation, procurement safeguards, consumer protections, and limits on high-risk uses. This emerging pattern makes AI governance not only a technology policy issue, but also a federalism and implementation issue. Depending on where people live, they may face very different rules governing automated decision-making in areas such as health, benefits administration, education, and employment. The resulting variation raises an urgent policy question: does state divergence create meaningful protection for residents, or does it produce an uneven geography of accountability and risk? This paper asks: How are states diverging in their regulation of public-sector AI, and what do those differences imply for equity, implementation capacity, and public accountability? The paper draws on the National Conference of State Legislatures AI legislation tracker; enacted statutes, executive orders, and agency guidance from selected states, including Colorado, California, New Hampshire, and Alabama; state procurement and oversight documents related to AI in government; and sector-specific public materials from health, education, and workforce policy settings where AI-related governance is explicitly addressed. States were selected through purposive comparative case selection to capture variation in public-sector AI governance approaches, including stronger protective models, administrative-management approaches, and lighter-touch or deregulatory frameworks. Using a comparative state policy design approach, the paper develops a typology of state AI governance models, distinguishing between stronger protective approaches, administrative-management approaches, and lighter-touch or deregulatory approaches. It uses document analysis to compare regulatory scope, enforcement provisions, transparency requirements, procurement safeguards, and protections for affected populations. Particular attention is given to whether state frameworks address equity harms, citizen recourse, and implementation capacity. Preliminary analysis suggests that state AI governance is diverging along multiple dimensions at once, including transparency requirements, procurement controls, restrictions on high-risk uses, and the availability of redress mechanisms. Some states are building more explicit protections around accountability and public oversight, while others are emphasizing flexibility, innovation, or agency discretion. Early findings indicate that this variation is likely to produce unequal protection across populations and policy domains, with important implications for how risks and administrative burdens are distributed. The paper argues that AI governance in the United States should be understood as a federalism problem with direct consequences for social equity and public administration. State variation is not merely a legal or institutional curiosity; it shapes who is protected, who bears risk, and how public institutions operationalize accountability. For APPAM audiences, the study shows how emerging AI governance regimes are producing a new map of unequal protection across the American state.

Author

• Albert Nii Noi Okwei, Virginia Commonwealth University
  Presenting Author