Governing the Ungoverned: Agentic AI and Accountability Vacuums in U.S. State Public Services

• In Event: AI Governance and Public Capacity in an Uneven Policy Landscape

Abstract

Background: As the federal government's role in domestic policy shrinks, state governments are assuming greater authority not only over traditional social programs but over the AI systems increasingly embedded within them. Agentic AI—systems capable of autonomous, multi-step decision-making with direct consequences for citizens—is being deployed across state public services including benefits administration, child welfare screening, workforce development, and tax enforcement. Unlike conventional algorithmic tools, agentic systems act without continuous human direction, compress timelines for intervention, and obscure chains of responsibility in ways existing governance frameworks were not designed to handle. Yet states are deploying these systems at dramatically different rates, under different procurement arrangements, and with strikingly inconsistent governance infrastructure.

Research question: How does state-level variation in agentic AI deployment create differential accountability gaps in public service delivery, and what governance responses are emerging—or failing to emerge—across U.S. states?

Methods: This paper presents a systematic comparative policy analysis of agentic AI deployment across U.S. state governments. Drawing on publicly available procurement records, state AI policy inventories, legislative tracking databases, and government accountability reports, states are evaluated across three governance dimensions: attribution (can responsibility for AI-driven outcomes be assigned?), oversight (are meaningful human review structures in place?), and contestability (do citizens have viable redress pathways?). This framework surfaces patterns of governance adequacy and deficit across the fifty-state landscape and connects variation in governance infrastructure to variation in citizen exposure to unaccountable AI-driven decisions.

Findings: Preliminary analysis reveals substantial cross-state variation—not only in whether agentic AI is deployed, but in the institutional infrastructure surrounding that deployment. A small cluster of states have enacted algorithmic accountability legislation or procurement standards with meaningful oversight provisions. The majority have not, producing conditions where consequential AI-driven decisions affecting citizens' access to public benefits operate in governance vacuums: no designated accountability actor, no audit requirement, no contestation mechanism. This variation maps partially but imperfectly onto state fiscal capacity, partisan ideology, and vendor market concentration.

Implications: State-level variation in agentic AI governance is a federalism and equity problem. Citizens in states with weak AI governance frameworks face a structurally different relationship with government than those in states with stronger accountability infrastructure, even when accessing nominally equivalent federally funded programs. As federal AI governance capacity recedes, this variation becomes more consequential. This paper argues for a minimum federal accountability floor for AI deployment in federally funded state programs and offers a three-dimension governance framework as a tool for researchers and policymakers assessing where gaps are most acute.

Authors

• Kelvis Cyon Cherisa Gills, Florida State University
  Presenting Author

• Deshawn Dontae Cooke, Florida International University
  Non-Presenting Co-Author