What Seattle Fire's AI Problem Should Tell Every Chief About Governance

 

A Seattle Times investigation published June 14 should be required reading for every fire chief in this country.

Here's what the reporting found: the Seattle Fire Department has been using AI to help triage and divert 911 medical calls for over two years. A Denmark-based company called Corti listens to every incoming medical call in real time and prompts dispatchers to route certain callers to a nurse line instead of sending an ambulance.

According to the Times, residents weren't told. There was no public review. And when reporters asked SFD how they were measuring outcomes — the department couldn't answer.

Before I go further, I want to say something clearly: Seattle Fire is one of the most innovative and respected departments in the country. Their contributions to EMS protocols, cardiac arrest response, and systems-level thinking have benefited fire and EMS agencies across the country for decades. If they're running into governance challenges with AI adoption, it's not because they're careless — it's because this technology is moving faster than any department's playbook has caught up with. Including yours.

That's the real story here. And it's one every chief needs to pay attention to.

Before I get into what this means for your department, I need to establish something clearly, because I'm going to reference a death in this post and the timeline matters.

First, the nurse line itself — before AI was involved.

In 2022, a retiree named Pamela Hogan called 911. She was diverted to Seattle's nurse triage line. She waited over ten hours for an ambulance. She never got care. She was found dead in her apartment. Her estate is now suing.

AI live prompting didn't start in Seattle's dispatch center until December 2023 — a year after Hogan died. Her case isn't an AI story. It's a nurse line story. But it's the foundation you need to understand before the AI layer gets added on top of it.

The relaxed ambulance response standards that applied in Hogan's case are still in place for the nurse line today. And according to the Times' reporting, the AI-assisted diversion rate has increased since then. That's the context your department needs to be sitting with.

When Vendors Lead, Chiefs Follow

The Times reports that Corti has been in a relationship with Seattle Fire since before the pandemic. A 2019 data-sharing agreement let the company train its AI on Seattle's 911 calls. Over the years, the technology expanded from quality assurance on completed calls to live prompts during active calls — on every medical call.

Here's a detail from the reporting that every fire chief should flag: Corti issued a press release claiming their AI drove a 50% increase in calls diverted to the nurse line. SFD later told the Times the actual number was 32%. Corti noted the figure aligned with their protocols — but the public number and the department's internal number didn't match, and the gap went unnoticed until a reporter asked.

That's worth sitting with. If your department is relying on vendor-reported metrics to evaluate an AI tool, you need an independent way to verify those numbers. Press release figures and operational outcomes are not always the same thing.

The Governance Gap Is the Real Problem

According to the Times' reporting, Seattle has a surveillance ordinance that requires review of technologies that observe or analyze individuals in ways likely to raise civil rights concerns. SFD has not confirmed that its use of Corti ever went through that review process.

Your jurisdiction may have similar requirements. More likely it doesn't — which means the governance gap is potentially wider. Either way, if you don't have a written framework for AI adoption in your department, you don't have governance. You have assumptions.

There's also a consent question unique to 911 that's worth raising with your leadership team: callers can't opt out. When someone is having a medical emergency, they call 911. There is no alternative. That raises the accountability bar compared to other AI use cases where people have a choice about whether to engage.

Is your city attorney aware of how AI is currently being used in your 911 system? That's a question worth asking — not because the answer is necessarily alarming, but because it's the kind of question that should have a clear answer before something goes wrong.

Admin vs. Operational AI — This Distinction Matters More Than People Realize

Here's something I want to say directly, because it gets lost in the broader AI conversation: not all AI use in the fire service carries the same risk profile. Not even close.

When AI helps a battalion chief draft an SOG, review a training plan, summarize an after-action report, or organize a budget narrative — the human is still the decision-maker at every step. You read it. You edit it. You approve it. If the AI gets something wrong, you catch it before it goes anywhere. The cost of an error is low, and the human review layer is built into the workflow by default.

Operational AI is a different category entirely.

When AI is influencing real-time dispatch decisions — who gets an ambulance, how fast, under what priority — the human review window compresses to seconds. Dispatchers are managing multiple calls simultaneously. A pop-up prompt in that environment isn't just a suggestion; it shapes behavior whether the designers intended it to or not. The cost of an error isn't a document that needs revision. It can be a patient who waits too long.

That's not a reason to avoid AI in the fire service. It's a reason to be deliberate about where you introduce it and what governance structures are in place before you do.

The safest place to start — and the place where the ROI is most immediate and measurable — is administrative work. SOGs, AARs, training documentation, grant writing, scheduling, email drafting, budget narratives. These are high-volume, time-consuming tasks where AI can save your staff hours every week, where errors are catchable before they matter, and where a chief can build AI literacy in their department without putting any operational outcome at risk.

Build confidence there first. Let your people get comfortable with the technology in a low-stakes environment. Then, when a vendor shows up pitching AI for dispatch or patient triage, your team will have the baseline knowledge to ask the right questions — and the credibility to push back if the governance isn't there.

This Is Not Staying Contained

The Times reports that SFD's own medical director said publicly last year that the next step is turning the AI on for higher-acuity calls. Meanwhile, 911 agencies in Snohomish and Kitsap counties are already using AI agents to answer non-emergency lines, with Snohomish also running an AI co-pilot on emergency calls.

This isn't an experiment in one city anymore. It's a direction.

The question isn't whether AI is coming to your dispatch center or your department's operations. It's whether you're going to be driving that decision or reacting to it after the fact.

What Chiefs Need to Do Right Now

This doesn't require a tech background. It requires the same instinct you already use for equipment and protocol decisions: Who approved this? What are the standards? Who's accountable if something goes wrong?

Start here:

1. Find out what's already running. Talk to your IT staff, your dispatch center, your software vendors. Ask directly: does this product use AI or machine learning? You may be surprised what's already active in systems you use every day.

2. Know where AI sits in your decision chain. There's a difference between AI that helps draft an SOG and AI that's influencing dispatch decisions on medical calls. The governance requirements for those two things are not the same. Map it before someone else does.

3. Start with administrative AI — on purpose. If your department isn't using AI yet, start where the risk is lowest and the value is highest: documentation, training plans, after-action reports, grant writing. Build the muscle before you're asked to make a decision about operational AI under pressure.

4. Verify vendor numbers independently. If a vendor tells you their tool improved outcomes by X percent, ask how they measured it and request the methodology. The Seattle situation is a useful reminder that public claims and operational data don't always match.

5. Get a written policy in place. If you don't have one, you're operating on assumptions — and assumptions are a poor foundation when accountability questions arise.

6. Ask whether your department has completed any required technology reviews. Many jurisdictions have surveillance or technology ordinances that may apply to AI tools. It's worth finding out before a reporter does.

The Opportunity

The chiefs who understand what's happening right now are the ones who are going to shape how AI gets used in the fire service — rather than having it shaped for them by vendors, city IT departments, and budget pressures.

Seattle Fire is an exceptional department that has led the fire service in technology and innovation for years. The EMS advances that came out of Seattle have saved lives in cities and counties across the country that have never heard of Corti. The fact that they're navigating these governance questions publicly is actually useful — it gives every other department a roadmap for what to get right before it becomes a headline.

That's the opportunity in front of you. Use it.