CENTERVILLE—In a sign of the times, the Centerville City Council recently spent close to an hour discussing implementation of a city personnel AI policy which would regulate its use among city staff.

The discussion, led by City Manager Brant Hanson on Oct. 21, highlighted the city’s need to provide guidelines for platforms like chatbots and text generators that are rapidly becoming common in the modern workplace. While Hanson and his staff argued the policy was essential for protecting the integrity of city information, council members were divided on how restrictive the new guidelines should be.

City council members agreed that a framework was necessary rather than a blanket approval of all AI tools. 

“I think that it is probably incumbent upon us to not just say all AI should be allowed,” Councilmember Brian Plummer said. “I think we should decide which ones we want to allow for the use.”

Councilmember Gina Hirst, who uses AI in her own work, noted its potential to boost efficiency, particularly for a city with a “lean staff and lots is expected.” She finds the technology helpful for tasks like drafting and editing.

“There are some moments where I’m finding it to be a really incredibly helpful tool to clean up a letter or help me to make sure it’s using more active verbs,” she said. “But I agree, I’m not sure we should be just cart blanch saying everything is allowed.”

However, some council members expressed concerns over AI's potential for inaccuracy, a phenomenon known as “hallucination.” 

Councilmember Cheylynn Heymann shared a sobering example she heard about at a judicial conference:

“AI sources are trained to give us answers that we want, right?” she said. “They are trained to please us in their answers... and a lot of times they will create fake information…Even a judge recently cited a made-up case in a judicial opinion, right? Like it is happening.”

Hanson said he had encountered this himself when his own research on county boundary adjustments yielded an answer that was “absolutely wrong” about the legal process. 

“Boy, if you’re going to be using ChatGPT or an AI... man, double check to make sure that that stuff is actually right because it may just be hallucinated,” Hayman said.

The risk of exposing sensitive city and resident information was then discussed. Hanson recounted what he called an alarming example where a county manager used an AI to draft a disciplinary letter containing confidential employee data. Months later, the private information resurfaced unexpectedly when another employee used the platform.

“That’s the scary thing about this whole thing,” Hanson said. “That’s why we’ve got to be so careful.”

Concerns were raised about the proposed policy giving the city’s IT manager too much power by granting them sole discretion to approve AI tools. Some council members also expressed a desire for a proactive list of approved, safer AI tools, rather than the reactive “Violations” section.

City Attorney Lisa Romney, however, argued for flexibility, citing the rapid evolution of the technology.

“To say that we’re limited to what we’re using right now is going to tie our hands and hamper our efficiency,” she said.

Hanson clarified that the policy’s intent is to create a framework that allows staff and an internal committee to vet and manage the tools, recognizing that “A lot of employees are using different tools depending on their job.”

Ultimately, the council voted 4-1 to approve the new AI policy, which will take effect on Dec. 15.

In the same meeting, the city council approved the formation of a data privacy committee. This new body will ensure the city complies with the recently readopted Utah Government Data Privacy Act, a law aimed at safeguarding citizens’ private information. The city is required to fully implement its data privacy program by Dec. 31.