IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

What If AI Didn't Need to Touch Government Data at All?

A recent example from the National Association of State CIOs Midyear Conference showed how, for some use cases, government might be able to skirt some of the privacy concerns surrounding generative AI entirely.

Scott Drzyzga, executive director of the Pennsylvania State Geospatial Coordinating Board, standing behind a podium and gesturing with both arms while speaking at the 2026 National Association of State CIOs Midyear Conference in Philadelphia.
Scott Drzyzga, executive director of the Pennsylvania State Geospatial Coordinating Board, presenting at the 2026 National Association of State CIOs Midyear Conference in Philadelphia.
Photo by David Kidd
When government works with AI, one of the fundamental concerns has been loss of data control — if AI uses inputs to learn, then how can an agency put sensitive or even confidential information into it?

Toward that end, many government agencies have been explicitly buying enterprise versions of AI platforms that don’t feed inputs back into the larger public model for training. In California, one agency’s guidelines to staff pointed out that nothing that could be subject to a public records request should be run through generative AI.

But what if there was a way to use AI without it even seeing the data? That could, essentially, eliminate those concerns about privacy and loss of control.

At the recent National Association of State CIOs Midyear Conference in Philadelphia, one example emerged of a local government doing exactly that. Scott Drzyzga, executive director of Pennsylvania’s State Geospatial Coordinating Board, showed off an example from Crawford County in the northwestern part of his state.

Crawford County recently added an AI chat feature to its parcel viewer. The parcel map is a common online fixture among county governments, allowing citizens and businesses to look up information about specific properties without personal assistance.

During the conference, Drzyzga demonstrated how the chat feature works with a prompt. First he selected a specific property, then said: “find properties within a half-mile that sold for more than $1 million in the last three years.”

“That’s a compound query,” he said. “I need to know where it is, I need to know how much it sold for, and I need to know when.”

The data was returned as a table, with its data plotted out on a mini map to the right. And the AI, he explained, was involved in a very narrow way.

“The AI agent never touches the data … it’s designed for a very limited use case,” Drzyzga said. “All it does is interpret the plain language question into SQL, it turns how people write and speak into code, and then that code is handed off to the next system. So the AI never remembers it, it never sees the owner information, it never sees the address, it never sees the value, it never makes a story, it never stores [data].”

In this case, AI is being used at the user interface layer to enable anybody to work with code regardless of their technical ability. So even if the core functionality of the parcel viewer application remains essentially the same, AI has opened up new uses for the website that many people never would have realistically attained on their own.

Do you have other use cases where AI improved something without needing access to data? Send them to Ben Miller at bmiller@erepublic.com.
Ben Miller is the associate editor of data and business for Government Technology. His reporting experience includes breaking news, business, community features and technical subjects. He holds a Bachelor’s degree in journalism from the Reynolds School of Journalism at the University of Nevada, Reno, and lives in Sacramento, Calif.