US public agencies report AI benefits—but only half describe unified governance
An EY survey of 200 state and local government decision-makers found widespread self-reported AI adoption alongside concerns about shadow tools, skills and funding. It measures leaders' perceptions, not audited deployment performance or public outcomes.
The short version
- What happened
- EY surveyed 200 US state and local government decision-makers: 100 at state level and 100 in local jurisdictions serving at least 100,000 people.
- Who it affects
- Public servants need task-level rules for approved tools, sensitive data and human review. Residents need evidence that any AI-assisted service improves access or accuracy without weakening privacy, appeal rights or accountability.
- What remains unproven
- Company-commissioned survey
Keep in mind
Company-commissioned survey
EY commercial interest
200 respondents
Self-reported adoption and benefits
No audited system inventory
No measured service outcomes
No public response rate
No questionnaire supplied in the release
Plus or minus 7-point total-sample margin of error
State and larger local jurisdictions only
Source analysis · 1 linked source. Primary analysed source date: 29 October 2026.
Read the full report · 3 min The evidence, methods, implications and limitations
Adoption is moving faster than a single agency-wide rulebook
US state and local government leaders told EY that generative AI is already moving into public administration, but oversight is uneven. In the online survey, 69% said their agency had fully integrated or was implementing generative AI. Among agencies that had begun some level of adoption, 96% reported at least one benefit: 59% selected greater efficiency, 56% improved cybersecurity or threat detection and 47% reduced administrative burden. Those answers describe leaders' assessments; the study did not inspect systems, service quality, error rates or savings.[1]
Shadow use and readiness remain the practical pressure points
Formal policy was common but not always unified. Ninety-four per cent said their agency had an AI governance policy or strategy, while 51% described it as comprehensive across the agency. Eighty-four per cent were concerned about unauthorised or unmonitored staff tools, 89% about securing dedicated AI funding and 86% about present-day skills, infrastructure and data readiness. For public servants, the immediate issue is less whether a policy exists than whether teams know which tools may handle citizen, personnel or operational data—and who reviews consequential outputs.[1]
The distinction that matters
What the survey can—and cannot—show
The findings are reported perceptions from decision-makers, not direct measurements of public services.
Sample
100 state + 100 local decision-makers
Leadership snapshot
Reported adoption
69% integrated or implementing generative AI
Self-reported status
Governance gap
51% comprehensive; 84% concerned about shadow use
Oversight concern
Not measured
No system or service audit
Performance unproven
A useful snapshot, not an independent performance audit
M Booth Research & Insights fielded the survey from 23 July to 4 August 2026. The sample split evenly between state decision-makers and local or municipal decision-makers from jurisdictions with at least 100,000 constituents. EY reports a plus-or-minus seven-point margin of error at 95% confidence for the total sample. The release does not provide response rate, recruitment frame, question order or subgroup uncertainty, and EY sells government modernization and AI services. Confidence would rise with a public questionnaire, independently sampled agencies, audited use inventories and measured changes in cost, accuracy, access, privacy incidents and staff workload.[1]