AI Reached Local Governments Before Policy Did

Elise Silva, PhD is the Director of Policy Research at Pitt Cyber; Beth Schwanke, JD is the Executive Director of Pitt Cyber; and Joy Ruff, AICP is the Executive Director of Local Government Academy.

Local government is the layer of government Americans encounter most often, running everything from public safety to parks to permitting. Yet the vast majority of research and policy action on artificial intelligence targets federal, state, or large city governments. Consider the Commonwealth of Pennsylvania, for example: of its 2,555 municipalities, nearly 90 percent have fewer than 10,000 residents. These are the boroughs, townships, and small cities where a handful of employees manage everything from procurement to payroll. Small governments touch millions of Americans' daily lives but are far harder to study than their larger counterparts.

This gap matters. The policy frameworks and guidance now taking shape assume staff and budgets that these small governments just don’t have. For the past year, the Institute for Cyber Law, Policy, and Security (Pitt Cyber) at the University of Pittsburgh has partnered with the Local Government Academy, a nonprofit that provides nonpartisan education to local governments in Pennsylvania, to find out how AI is being used in small government settings.

The 35 respondents to our recent survey, mostly senior staff from townships and boroughs serving 5,000 to 20,000 residents in Pennsylvania, offer a small, self-selected sample. Their responses do not present a national picture but do provide meaningful ground-level data on a population few studies are reaching, offering clear direction for how policy researchers can help smaller local governments implement AI responsibly.

Among our respondents, AI use is already common, but basic governance hasn't caught up. Two-thirds said they use generative AI tools at least monthly for work. More than half reported their workplace had rolled out enterprise tools like Microsoft Copilot, yet more than a quarter said they were also using personal, free accounts like OpenAI’s ChatGPT for work tasks (a detail that should concern anyone thinking about data security in government settings). Despite that level of use, the majority of respondents have no genAI policy at all, yet more than half of those without a policy said it would be valuable to have one. As these responses show, the demand for guardrails is real.

Leadership attitudes may help explain the lag. About one fifth of the respondents said their leadership was actively encouraging AI exploration and adoption and another quarter said their leadership was supportive but not actively promoting AI use. Around two-fifths–the largest group–described their leadership’s position as neutral or unclear, and only two respondents to this question said leadership actively discouraged use. A similar number reported conflicting signals from their leadership. As in other sectors like education, AI arrived rather than being formally adopted. It is often slipping in through the cracks and is being used alongside already-established workflows, rather than being put in place to intentionally disrupt those workflows or being directly procured. While using AI may be as simple as opening a browser, governing it responsibly requires substantially greater organizational capacity. That reality, compounded by the absence of policy resources, may be why leadership sits in neutral. AI has spread differently from earlier technologies, most often arriving through individual use rather than institutional decision.

So, what are people doing with these tools? As of the summer of 2026, mostly writing, summarizing, and taking meeting notes. Well over half cited these productivity tasks as their primary use. Notably, a handful of the respondents mentioned surveillance and identification. While a minority, with the possibility of no policies in place and personal accounts in place, their concern is worth noting.

Asked what excites them about AI, respondents pointed to efficiency gains, data analysis, and accessibility improvements. But they were equally vocal about what worries them: data privacy, loss of human connection, confidentiality, bias, and legal exposure all came up in open-ended responses. The trade-offs between the gains and the risks associated with these tools were well recognized by local government employees.

And what do these employees need given these realities? One finding was unambiguous: they want training and support to use and govern these tools. Half of the respondents cited a lack of staff expertise as the biggest barrier to responsible adoption, ahead of budget constraints at thirty-eight percent. Training, policy guidance, and data security concerns followed closely behind. When asked directly what kind of support would be most useful, respondents were clear: they wanted to see how peer governments in their own region are using AI and basic training on using these tools for everyday office work. A formal report or written guidance ranked well below live peer-to-peer learning. Local governments do not want another framework. Instead, they want to talk more to each other about what worked, what didn't, and why.

This does not contradict the case for policy research and artifacts; rather, it clarifies what would be most useful for smaller governments. Much of the current conversation about public-sector AI—procurement standards, model evaluations, pilot programs in major cities—assumes a level of staffing and budget that smaller governments simply do not have. Our survey suggests that what these governments want is more basic and more local. They would like help writing a use policy, straightforward training, and examples from governments that look like their own. One way to meet this need is activating networks that are already positioned within civil and public society, including state municipal associations, councils of governments, universities, and regional nonprofits like the Local Government Academy. These organizations are positioned to translate best practices into locally applicable guidance, and to convene the peer networking and collaborative learning that respondents asked for.

Small jurisdictions like those represented in our survey make up the overwhelming majority of American governments, yet they get the least attention in AI policy and governance. Our data—while limited—shows use running ahead of rules to guide it. We worry that if adoption keeps outpacing governance, the potential costs from compromised resident data to biased outputs to eroded public trust, will land on the governments least equipped to absorb them.