Jefferson commissioners work with AI draft policy

Published 1:30 am Saturday, May 23, 2026

PORT TOWNSEND — The Board of Jefferson County Commissioners considered a draft of an artificial intelligence policy during a workshop.

In spite of public critiques of the artificial intelligence (AI) policy on Monday, Commissioner Greg Brotherton said the draft represented progress.

“I think this is the strongest draft policy we’ve had so far,” he said.

The policy addresses the security and containment of sensitive data, the blocking of unauthorized platforms, and standards for public attribution and transparency when using AI tools.

Developers at Darwin AI, the software used by Jefferson County, provided feedback on the draft to the prosecuting attorney’s office, Central Services Director Shawn Frederick said.

The prosecuting attorney’s office also ran the policy through a Darwin AI policy development tool, he added.

“What you see here is a combination of the best parts of those three efforts: Having a policy reviewed by AI professionals, creating an AI policy using the AI tool within Darwin and, of course, the policy work that we’ve been undertaking now for the better part of two years,” Frederick said.

The policy is complex for several reasons, Frederick said.

“The technology is very new, and it is quickly becoming very broad,” he said.

The purpose of the policy is multi-layered. It is to provide governance on how AI is being used at the county and it is aspirational.

“It is to drive globally our collective ability to use AI effectively,” Frederick said.

Frederick highlighted the first value in a section on values and guiding principles.

“The one that is most unique to AI is ensuring there’s a human in the loop, that there isn’t an opportunity or a direction that takes us down a course where we’re using AI to make significant or consequential decisions,” he said. “We also want to make sure that all of our technologies are safe and secure.”

Some AI systems already have been marked as unauthorized based on their geographical origin.

“Specifically, AI tools that originate out of China, we’ve automatically identified those tools and essentially blacklisted them,” Frederick said.

As of Monday, Frederick said there were about 80 AI tools and agents existing within the county’s system. Many of the tools are those that exist within more general softwares used by the county. For example, Microsoft Copilot is in Office 365 and Gemini is in Google.

“We have AI systems in many of the systems we use every day that we wouldn’t necessarily recognize as AI,” Frederick said.

Frederick said the county’s initial use of generative AI tools centers around Microsoft Copilot, which is already approved.

“We have the most amount of control over that system,” he said. “Because of our licensure of that system we implemented last year, we know that our data remains with us and it doesn’t go outside of the agency to feed a large language model, which is important until we validate how that data is retained, preserved, our accessibility to it in terms of record retention, public records request and all of those things.”

Jefferson County resident Tom Thiersch shared what he found by running the policy through ChatGPT during public comment.

“I submitted your proposed policy to ChatGPT, and its analysis was rather interesting,” Thiersch said. “ChatGPT summarized and said, ‘Well, as an operational government document, it’s over engineered, difficult to administer, internally inconsistent in places and likely too complex for effective countywide implementation without substantial simplification and supporting procedures.’”

Thiersch also noted issues with the scope, including people who volunteer for the county. Thiersch, who has volunteered for the county for 20 years, said he likes using AI tools other than Copilot.

“In fact, I used ChatGPT to write some language for a ballot supporting the county’s transportation benefit district sales tax,” he said.

Chief Civil Deputy Prosecuting Attorney Philip C. Hunsucker said the policy’s rules would only restrict activity occurring on county infrastructure.

“The whole policy is limited to how we use AI on our equipment,” he said.

The commissioners addressed the public’s concerns regarding the document’s language — specifically a section that repeated the words “we expect” repeatedly.

Hunsucker said the language was used deliberately to distinguish aspirational targets from hard rules.

The stakes of violating the policy are outlined in the accompanying workshop documents, which state that violations of the established guardrails carry steep disciplinary penalties. Those actions include the denial or revocation of access to AI tools, the termination of third-party vendor contracts, direct personal liability for harms caused by negligence, and formal termination of employment.

The board spent a portion of the workshop unearthing errors within the draft, noting, among other things, it inconsistently swapped titles like “AI Working Group” and “AI Review Committee.”

“I feel like we need to bring it back before we start doing an actual line-by-line of the edit,” Commissioner Heather Dudley-Nollette said.

A further-refined policy is likely to return to the board on June 8.

“We’re in this limbo where we don’t have a policy to govern how these tools are being used,” Frederick said. “At the end of the day, the security of the system is my responsibility. The only tool I have today is to block it. That’s not what I want to do. I’m a huge AI proponent.”

________

Reporter Elijah Sussman can be reached by email at elijah.sussman@peninsuladailynews.com.