The Ghost Officer
The King County Prosecuting Attorney's Office drew a line in the sand. Here's what the Ghost Officer problem means for your reports and your career.
A Memo That Changed the Game
In 2024, the King County Prosecuting Attorney’s Office (KCPAO) sent a formal memo to all law enforcement partners in King County.
The subject: AI-generated incident reports.
The message: We will not accept them.
Not “we have concerns.” Not “we’d like additional review.” A flat refusal — and the reason they cited has a name.
The Ghost Officer.
What Is the Ghost Officer?
The Ghost Officer is the KCPAO’s term for a specific, dangerous type of AI hallucination:
The risk of an AI generating observations, statements, or narrative actions that the physical officer on the scene never actually witnessed or performed.
A Ghost Officer is an AI-invented version of you — appearing in your own report — doing things you never did, seeing things you never saw, recording statements that were never made.
The Ghost Officer isn’t you. But your signature is on the document.
Why This Destroys Cases
When you sign an incident report, you are legally attesting to its accuracy. You are saying: I was there. I witnessed this. This is what happened.
If that report contains actions, observations, or statements that a Ghost Officer performed — not you — you have signed a false legal document.
Here’s what follows:
Brady violation: Prosecutors are required to disclose any information that might be favorable to the defense. A report known to contain AI-generated content that wasn’t verified is discoverable as a potential Brady issue — meaning the defense gets it, and gets to argue it undermines the case.
Impeachment on the stand: Defense attorneys are exceptionally good at cross-examination. If they can demonstrate that your report contains events you can’t personally confirm from your BWC or memory, your credibility as a witness is damaged — not just in this case, but in every future case you testify in.
Wrongful charge exposure: If the Ghost Officer’s invented facts form the basis of a charging decision, you may be looking at a case that gets dismissed, overturned, or — in serious civil rights contexts — a lawsuit.
The SPD Officer Who Admitted It
In Seattle, an officer publicly acknowledged using AI-generated content in an incident report. The Seattle Office of Police Accountability (OPA) responded by urging SPD to develop a formal AI use policy.
The MyNorthwest headline at the time: “After officer admits to using AI, Seattle OPA urges SPD to create policy.”
The officer wasn’t disciplined for using AI. The problem was using it without appropriate disclosure and verification. The policy gap that moment exposed is exactly what this course addresses.
Read each scenario. Rate the Ghost Officer risk: Low / Medium / High — and explain why.
Scenario A: Officer uses Code Four to draft a report for a simple trespassing call. One individual, no dispute, officer’s BWC captures the entire interaction clearly. Officer reviews the draft line by line against the footage.
Scenario B: Officer uses Draft One after a chaotic crowd disturbance involving 12 individuals, multiple use-of-force incidents, and 40 minutes of overlapping audio. Officer skims the draft and makes a few small edits.
Scenario C: Officer uses Draft One for a domestic disturbance. The suspect made several verbal statements the officer can’t clearly remember. The AI draft includes detailed quotes from the suspect. Officer assumes the AI captured them correctly.
Answers:
A — Low risk. Simple scene, clear BWC audio, thorough review. The human-in-the-loop process is working correctly.
B — High risk. Complex scene with many voices, high hallucination probability, and insufficient review. Any Ghost Officer content has multiple opportunities to slip through.
C — High risk. The officer is relying on the AI to fill memory gaps — exactly the wrong use of the tool. Verbal statements are among the most frequently hallucinated and most legally consequential elements. Every direct quote must be verified against BWC audio.
You won’t. Not reliably.
Memory is reconstructive, not playback. After a high-stress call, your brain will fill gaps with what you expected to happen, not necessarily what did. When you read an AI draft that sounds generally right, your memory will tend to confirm it — even if specific details are wrong.
This isn’t a weakness. It’s how human memory works. It means that trusting your memory to catch AI hallucinations is not a verification strategy. BWC footage is your verification strategy.
An AI-generated report quotes a suspect as saying: 'I know I shouldn't have taken it.' The officer has no memory of the suspect saying this. The BWC audio from the scene is muffled and the statement can't be confirmed. What should the officer do?
Reveal Answer & Explanation
Remove the quote from the report and document what actually occurred.
If you cannot confirm a direct quote through BWC footage, witness accounts, or your own clear recollection, it does not belong in a signed incident report as a direct quote.
Options:
- If the suspect made a statement you can’t precisely recall, write: “Suspect made statements indicating awareness of the alleged offense” — a paraphrase that doesn’t overstate certainty
- If you genuinely can’t confirm any statement was made, document that no statement was given
- Do not sign a report containing a direct quote you cannot personally verify
The Ghost Officer risk is highest in exactly this scenario: when the AI fills in statements the officer can’t confirm. This is how reports containing false confessions, phantom admissions, and invented narratives get signed and filed.
The Bottom Line
The KCPAO memo isn’t a bureaucratic overreaction. It’s a legal system drawing a clear boundary: reports signed by humans must reflect human-verified facts.
AI can draft. AI can organize. AI can save time.
But the signature belongs to you — and so does the legal accountability for every word it covers.
In Lesson 3, we look at the full risk picture: the matrix of what every AI tool gains and what it puts at risk — side by side.