The deposition room was quiet except for the sound of the defense attorney flipping through pages. Officer Marcus Webb had been on the job for eleven years and had testified in dozens of cases. He knew what a deposition felt like. This one was different. The attorney looked up and said: "Officer Webb, this report. Did you write it, or did a computer write it?" Webb paused. He had used Axon Draft One for about seven months. The draft had been generated from his body-worn camera (BWC, the recording device officers wear to document encounters) audio. He had reviewed it, corrected one sentence, changed a time by three minutes, and submitted it. "I reviewed it," he said. The attorney placed a specific sentence on the table: a description of the suspect's location and posture at the moment of the officer's approach. "Did you write this sentence?" Webb looked at it. He was not certain. He believed it was accurate. He thought he remembered it matching what he had seen. But he could not say with certainty whether it came from his own observation or from the AI draft that he had accepted without checking that particular sentence against the footage. He said: "I reviewed the report." The attorney said: "That's not what I asked." That gap, the space between "I reviewed it" and "I wrote and verified every factual claim in this report," is the gap this lesson is about. Closing it is the entire professional obligation of the cardinal rule.

What the Cardinal Rule Is

The cardinal rule of AI-assisted report writing in public safety is this: you are the author. Not in the sense that you typed every word. You may not have typed any words. The AI may have generated the entire narrative from the body-camera audio, and your contribution may have been to correct a time and fix a spelling error. That is fine. That is the point of the tool. The cardinal rule is not about who produced the keystrokes. It is about who is accountable for every factual claim in the document, who verified every factual claim against the record, and who will stand behind every factual claim in a deposition, a cross-examination, and in front of a jury.

That person is you. Not the vendor. Not the tool. Not the department that deployed the tool. Not the supervisor who approved the report. You: the officer whose name is on the document, whose badge number is in the header, and who submitted the report as their sworn account of what occurred.

This is not a new principle. Sworn reports have always required that the officer behind them be accountable for their accuracy. What the AI drafting tool changes is the practical mechanics: an officer who typed a report from memory and notes was, by the act of typing, putting their own words into the document. They may have misremembered, and memory errors in reports are real and documented, but the report was theirs in a direct sense. An officer who adopts an AI draft has received a document generated by a system that does not observe, that does not remember, that fills gaps from statistical patterns, and that cannot distinguish its accurate outputs from its invented ones. The accountability is the same. The process of establishing that accountability requires an additional step: the footage-grounded verification pass that is described in the next lesson in this program.

The Deposition Question and Why It Matters

The question Officer Webb could not fully answer, "did you write this sentence?," is not an unusual question. It is becoming a standard question in jurisdictions where AI drafting tools are in use, and defense attorneys are learning to ask it. The question is designed to surface the gap between passive review and active verification: between an officer who looked at a report and confirmed it seemed right, and an officer who checked each factual claim against the footage and can say, "I verified this sentence against the body-camera recording at timestamp 4:22, and this is what the footage shows."

Under the oath an officer takes when testifying, "I reviewed the report" is a statement about process, not about accuracy. It does not tell the court anything about whether the factual claims in the report were verified against the evidence. A defense attorney who hears "I reviewed it" will follow up with "What did the verification consist of?" and "Can you tell me which parts of the report came from the audio transcript and which parts were generated by the AI based on its training patterns?" Those follow-up questions reveal, quickly, whether the officer performed substantive verification or passive acceptance.

The deposition answer that closes the gap is specific. It is not "I reviewed the report." It is: "I generated a draft from the body-camera footage using the department's AI drafting tool. I then reviewed the draft against the BWC recording and the CAD (computer-aided dispatch, the system dispatchers use to log calls and track incident details) entry. I verified each factual claim against the record. Where the draft had details I could not verify against the footage, I removed or corrected them. I adopted the final version as my own sworn account." That answer does several things at once. It discloses AI assistance. It documents the verification process. It establishes the officer as the author of the verified, adopted account. And it closes the door on the "the computer wrote it" challenge before the defense attorney can open it.

Why "The Computer Wrote It" Is Never an Acceptable Answer

It is worth stating clearly why "the computer wrote it" is not a legally or professionally acceptable answer in any context: deposition, cross-examination, internal review, or commander briefing.

Legally, a sworn police report is the officer's sworn statement. The officer's signature on the report, or the electronic submission that constitutes the officer's adoption in the records management system (RMS, the agency's central database for case files and incident reports), is the officer's attestation that the contents are accurate to the best of their knowledge and belief. There is no clause in that attestation that says "except for the parts the computer generated." The signature covers the whole document. If the document contains an AI-generated fabrication, and the officer signed it without verifying that particular claim, the officer has attested to something they did not verify. The fact that they did not generate it does not reduce their accountability for attesting to it.

Under Brady v. Maryland (1963, requiring disclosure of evidence favorable to the defense) and Giglio v. United States (1972, requiring disclosure of impeachment evidence including evidence affecting a witness's credibility), the obligations fall on the human participants in the criminal justice system, not on the tools they use. A Brady violation does not have a "but the AI did it" exception. A Giglio credibility problem does not disappear because the inaccurate characterization came from an AI draft. The constitutional obligations are human obligations, and they are met or violated by humans, not by software.

Practically, "the computer wrote it" removes the officer from the role they are legally required to occupy: the author and verifier of their own sworn account. It transforms the officer from a professional who used a tool to produce a verified document into a conduit who passed an unreviewed AI output into the criminal justice system. The first officer is a professional. The second officer is a liability for the agency, for the prosecution, and for every case their reports touch.

What Authorship Actually Requires

Accepting that you are the author is the principle. The practice that gives it content is the review-and-adoption workflow. What does it actually mean to author a report that was initially drafted by AI?

It means, at minimum, the following. First, you must read the entire draft, not scan it. The gap-fill detail (the AI-invented detail that fills a transcript gap) and the softened fact (the AI characterization that modifies the legal significance of an event) will not announce themselves in a scan. They require deliberate reading.

Second, you must check every factual claim against the BWC footage and the CAD entry. Every attributed statement. Every description of a physical location or object. Every sequence of events. Every time. This is the footage-grounded verification pass. It is not a quality check in the sense of reviewing a document for coherence and clarity. It is an evidence verification: confirming that each claim in the report is supported by the footage, the dispatch record, or your direct observation at the scene.

Third, you must correct or remove every claim you cannot verify. The appropriate response to an AI-generated detail that you cannot confirm against the footage is to remove it from the report, not to leave it because it seems plausible. Plausible-but-unverified details are exactly the gap-fill details that become Brady problems. If you cannot say "I verified this against the footage at timestamp X," the claim does not belong in your sworn account.

Fourth, you must add what the AI missed. The AI draft is generated from the audio transcript. It cannot capture what you observed but did not verbalize, what the CAD entry contains that was not in the audio, or what other officers or witnesses reported that was documented separately. Your review is also an authorship step: you are adding the observations and facts that the AI transcript missed, because your sworn account must be complete, not just accurate about what the AI included.

Fifth, you must adopt the final document as your own. In most RMS deployments, this means the electronic submission with your credentials. Some agencies are adding an explicit AI-use disclosure notation: "This report was drafted with the assistance of AI from BWC audio. The reporting officer verified all factual claims against the BWC recording and CAD entry and adopts the final version as their sworn account." That notation is not just bureaucratic compliance. It is the documentation that allows you to answer the deposition question completely, specifically, and without hesitation.

The Agency Obligation Beside the Officer Obligation

The cardinal rule is primarily addressed to the individual officer. But individual officer accountability does not exist in a vacuum. The agency that deploys AI drafting tools has its own obligations that make the cardinal rule practically achievable rather than theoretically demanded.

An agency that deploys AI drafting tools without a verification protocol is asking officers to be accountable for a standard it has not equipped them to meet. If the department says "use the AI tool but no explicit verification step is required," it is setting up a situation where the officer's accountability is asserted but the mechanism for exercising it is absent. The outcome is exactly what happened to Officer Diallo in the previous lesson's opening story: an officer who did not know precisely what verification was supposed to look like, trusted a tool that had been consistently good, and ended up with an adopted report containing three hallucinations that dismissed a case.

The agency obligation includes: a written verification protocol that tells officers specifically what to check and against what source; a disclosure policy that specifies how AI assistance is to be documented in the report or in the case file; a training program that teaches officers to perform the footage-grounded verification pass, including how to catch gap-fill details and softened facts; a supervisor review process that confirms the verification was performed; and a CJIS (Criminal Justice Information Services Security Policy, governing the handling and transmission of criminal justice data) compliant data-handling framework for the AI tool's use of case-file data.

An officer who has been trained, given a specific verification protocol, and is operating within a disclosure policy has the tools to be the author in the full professional sense. An officer who has been handed an AI tool with a "use your judgment" instruction does not. The cardinal rule holds in both cases, but the professional obligation to actually enforce it rests on both the officer and the institution.

The Audit Trail and the Deposition Answer

The final element of making authorship real is documentation. An officer who has performed a thorough footage-grounded verification pass but has no record of doing so is in a weaker position than an officer who has both performed the pass and has a record of it. In a deposition, "I verified this" is better than "the computer wrote it." "I verified this, and here is the notation in the case file documenting the verification" is better still.

The most sophisticated agencies deploying AI drafting tools as of mid-2026 are building audit trails that log three things: that the report was AI-assisted (tool version, input, timestamp of draft generation), what the officer changed from the original draft (a change log), and that the officer reviewed and adopted the final version (an explicit adoption record). This audit trail is the documentation that supports the deposition answer. It allows the officer to say not just "I reviewed the report" but "here is the audit log showing what the AI generated, here is what I changed and why, and here is the adoption record documenting my review."

That audit trail also meets the King County prosecutor's requirement for documented verification, and it satisfies the Brady-Giglio framework by documenting that the process used to produce the report is transparent and accountable. It is the tool the officer uses to answer not just the defense attorney's question, but the questions a prosecutor, an oversight board, and a jury might ask. And it is the documentation that, decades from now, allows a cold case review to understand exactly how a report was produced.

It helps to follow a single sentence on its full journey, because that journey is what the cardinal rule is really about. Suppose the AI draft contains the line "the suspect appeared agitated and refused to comply." The model produced that sentence in under a second, assembled from audio patterns and the statistical shape of thousands of similar narratives. The moment the officer reads it, the sentence stops being a machine's guess and becomes a claim the officer must stand behind. Months later a prosecutor reads the same sentence and decides whether to file a resisting charge on the strength of it. A defense attorney reads it and pulls up the BWC footage to test whether "agitated" and "refused to comply" match what the camera actually recorded. A judge reads it when ruling on a suppression motion. A juror, if the case reaches trial, hears it read aloud and weighs it against the officer's demeanor on the stand. That one sentence, generated in a second, can travel for years through the most consequential machinery in government, and at every stop the only name attached to it is the officer's. The cardinal rule is simply the recognition that the speed of generation and the weight of consequence live in the same sentence, and that the officer is the bridge between them.

This is also why the cardinal rule does not soften as the technology improves. A more accurate model produces fewer errors, which is genuinely valuable, but it does not change who is accountable for the errors that remain. In fact, a more fluent and confident draft can make the verification pass harder, because a polished sentence invites the reader to trust it. The better the tool gets at sounding right, the more disciplined the human author has to be about checking whether it is actually right against the record. Accountability is not a temporary scaffold that the technology will eventually outgrow. It is the permanent foundation the entire workflow is built on, and it is the one part of the job that no model release will ever take off the officer's desk.

The computer wrote the draft. You wrote the report. The difference is the verification pass.

Key Takeaways

• The cardinal rule is: you are the author of your sworn report, regardless of what tool generated the initial draft. Authorship means accountability for every factual claim, not the production of every keystroke.
• "The computer wrote it" is never an acceptable answer in a deposition, a cross-examination, an internal review, or a commander briefing. The officer's signature on the report is an attestation to the entire document, not just the parts the officer typed.
• Brady v. Maryland and Giglio v. United States impose constitutional obligations on human participants in the criminal justice system. Those obligations have no "but the AI did it" exception.
• The review-and-adoption workflow is what makes authorship real in practice: read the entire draft, check every factual claim against the BWC footage and CAD entry, correct or remove unverifiable claims, add what the AI missed, and explicitly adopt the final version as your sworn account.
• The deposition answer that closes the gap between "I reviewed it" and "I am the author" is specific: it discloses AI assistance, documents the verification process, establishes the officer as the accountable author, and describes what was verified and how.
• An agency that deploys AI drafting tools without a verification protocol, a disclosure policy, a training program, and a supervisor review process is asking officers to be accountable for a standard the agency has not equipped them to meet.
• An audit trail that logs the AI draft, the officer's changes, and the adoption record is the documentation that supports the deposition answer, satisfies the King County standard, and meets Brady-Giglio transparency requirements.
• The professional distinction is between an officer who used a tool to produce a verified document and an officer who passed an unreviewed AI output into the criminal justice system. The cardinal rule makes you the first kind of officer.