Bonner County commissioners seek department use inventory, legal review before countywide AI policy

Bonner County commissioners directed staff on Oct. 12 to survey elected officials and department heads about current and planned uses of generative artificial intelligence, and to begin rolling out employee training while the county obtains a legal opinion on recordkeeping and disclosure.

The action came after a lengthy discussion about employee use of ChatGPT, Gemini and other AI tools, unresolved public-records questions and possible costs and privacy trade-offs. "If we're going to be transparent with the public ... maybe I provide that and say, this is how we're using it," one participant said, urging the county to be able to show logs of AI use if questions arise.

Commissioner John Williams (first reference: 0:24) framed the issue as one of scope and risk, saying departments already use AI for tasks from drafting to summarizing documents but that the county currently lacks a way to log or audit those inputs. Director of Technology Storms (first reference: 2:39) described technical options: assigning centrally managed accounts that token‑track use, running a local interface that still calls external AI servers, or hosting an in-house model at much higher hardware cost. Storms said training modules—identified as available in the county's learning system "Safety plus" and an earlier product "Knowl"—could be deployed to staff through HR.

Legal and public-records questions were a central concern. Multiple speakers noted uncertainty about whether AI prompts and other intermediate inputs create a disclosable public record distinct from finished work product. An outside prosecutor referenced by the group, Bill Billman Allen of Shoshone County, was described as advising that the final product (for example, a mailed letter) is what is ordinarily produced in response to records requests, while internal drafts are not always producible.

Cost considerations were raised: one example given was $20 per month per user for a commercial account; another illustration compared tokenized, pay-as-you-go models with sample rates cited at about $0.14 per million words into a service and $0.40 per million words out. Commissioners also discussed non-monetary costs—potential loss of control over proprietary or sensitive data if staff paste those materials into third-party services—and the need for technology controls to prevent unauthorized downloads or installations.

There was no vote; commissioners and staff agreed on a two-step approach. First, the technology director will ask each elected official and department head to report current and anticipated AI uses so the county can identify where more specific policy or training is required. Second, the county will deploy available AI-safety training modules through its learning platform and seek a legal opinion on whether prompts, inputs or intermediate AI records are subject to public-records disclosure. Participants said the results of the departmental survey and the legal guidance would be returned to the board for further policy work.

Commissioner Donke and Assessor Englehart urged acting sooner rather than later: Englehart encouraged a broad initial policy that technology can refine later, stressing liability and the need to train users before granting access. Several participants suggested that a countywide policy should allow for department-specific rules because certain elected offices control their own data and records.

Staff also noted practical controls that can be applied during a migration from existing services (for example, limiting or disabling assistant features such as Gemini or Copilot in corporate email) and the option to approve a short list of permitted AI services while blocking others.

The board did not adopt a final policy at the meeting. Instead, commissioners requested the department-use survey be issued to elected officials and department heads, asked technology staff to load AI safety modules into the county learning platform for employees, and requested that legal provide a formal opinion on public‑records and disclosure obligations for AI-generated content. The topic will return for review once those items are compiled.