The defense attorney for a man charged with aggravated assault takes his seat at the deposition table. Across from him sits the patrol officer who wrote the report. The attorney opens to page four of the narrative and asks, simply: "Officer, did you write this report?" The officer says yes. The attorney points to a sentence in the middle of the narrative: a quote from the defendant that describes his physical state at the time of the arrest. "Did you personally write this sentence?" There is a pause. The officer knows the draft came from the AI tool. The sentence reads accurately, more or less, but the specific phrasing, a direct denial with a specific clause about injury, is not something the officer can place in the footage. The officer says "I reviewed and adopted this report as my own." The attorney smiles slightly and says: "I see. So you adopted a report that contains a quote from my client that does not appear in your body-camera footage?" The deposition continues for another two hours. By the end of it, the question of whether that sentence was accurate is not the only issue. The question of how the report was produced, who actually wrote it, and whether every sentence in it was verified against the recording has become a central question in the case. This lesson is about why that scenario is not a hypothetical and why the single fact that a police report is evidence, not a memo, reorders every rule about AI in public safety.

The Document That Goes to Court

Most professionals who work with documents use AI to help draft documents that will be reviewed internally, revised multiple times, sent to clients who may not act on them for weeks, or filed for record-keeping purposes. In those contexts, a first draft with some inaccuracies is expected, caught in review, and corrected before it matters. The document is a working tool on the way to a final product. The stakes of an early draft being wrong are manageable.

A police report is not that kind of document. The moment an officer completes and submits a report, it begins a journey that is nothing like the journey of a business memo or a corporate summary. It is uploaded into the RMS (records management system). It is pulled by prosecutors preparing a charging decision. It is disclosed to defense counsel as part of discovery. It is read in depositions. It is read into the record at preliminary hearings. It is the basis for testimony. It is used by defense attorneys to cross-examine the officer. And in serious cases, it is part of the file that a jury sees, or at least the basis for testimony about what happened at the scene.

This trajectory is what makes AI assistance in report writing categorically different from AI assistance in almost any other professional context. The errors that survive a police report do not stay in the report. They travel into the legal system, and they create problems in places the officer may not expect, in a deposition eighteen months later, in a suppression motion hearing, in a Brady disclosure review by a prosecutor who flags a potential inconsistency between the report and the footage.

Every program and vendor that talks about AI-assisted report writing is, necessarily, talking about AI-assisted evidence production. That framing is not an exaggeration. It is a precise description of what is happening. The draft the model generates from the body-camera audio is a candidate for a document that can put someone in prison or set them free. That single fact is the reason this program has the standards it does, and it is the reason every lesson in this program returns to verification, authorship, and disclosure.

Brady and Giglio: The Constitutional Frame

To understand why AI-related errors in police reports are not just embarrassing or inconvenient but constitutionally significant, you need to understand two landmark cases that define the prosecution's disclosure obligations in the American criminal justice system.

Brady v. Maryland

Brady v. Maryland, decided by the Supreme Court in 1963, established that the prosecution must disclose to the defense any evidence that is material to guilt or punishment that is favorable to the defendant. This is called Brady material or exculpatory evidence. The rule is grounded in the Due Process Clause of the Fourteenth Amendment. It is not optional, it is not subject to prosecutorial judgment about what the defense "really needs," and it applies even if the prosecution did not intend to suppress the evidence. If the prosecution possesses evidence that would help the defense and does not disclose it, that is a Brady violation.

How does this connect to AI-drafted reports? The connection runs through the gap-fill detail and the softened fact. If an AI-drafted report contains a detail that is incriminating but was not in the footage, and the defense later obtains the footage through discovery and finds that the reported detail is not there, the prosecution is now in a Brady territory. The footage, which does not show the incriminating detail, is exculpatory material that the prosecution must disclose. But more directly: if the officer adopted a report containing a fabricated incriminating detail and the prosecutor relied on that report in making a charging decision, the prosecution may have proceeded on the basis of false evidence. That is a Brady problem with consequences that can extend to case dismissal, professional discipline, and in serious cases, civil liability.

The connection also runs in the other direction. If the AI draft softened a detail that was favorable to the defense, and the report therefore does not contain material that was present in the footage, the prosecution may be in a Brady violation for the material that the AI's softening omitted. The footage is the ground truth. Any discrepancy between the footage and the report is a potential Brady problem.

Giglio v. United States

Giglio v. United States, decided by the Supreme Court in 1972, extended Brady's disclosure requirements to include impeachment evidence, specifically evidence that bears on the credibility of prosecution witnesses. This includes information about a witness's prior misconduct, prior false statements, or anything else that would help the defense challenge the witness's credibility at trial. In the context of police officer testimony, Giglio material includes the officer's disciplinary history, prior findings of untruthfulness, and prior instances in which the officer's reports were found to be inaccurate.

The Giglio connection to AI is this: if an officer adopts an AI-drafted report without adequate verification, and that report contains inaccurate statements, and that pattern of inadequate verification is documented across multiple cases, that pattern is Giglio material. A defense attorney who discovers that Officer A has a documented history of adopting AI drafts that were later found to contain inaccuracies has material that bears directly on Officer A's credibility as a witness. That material must be disclosed. And if the prosecutor's office has that information and does not disclose it, that is a Giglio violation.

King County, Washington, became a widely cited example in 2025 and 2026 when the county's prosecuting attorney's office issued guidance barring AI-generated police reports from their courtrooms unless the reports met specific documentation and verification standards. The prosecutor's concern was exactly the Giglio problem: that AI-drafted reports without documented verification created impeachment risk for officers who testified about their reports, and that the office could not fulfill its Brady and Giglio obligations if it could not verify the provenance of every statement in the reports it used. The King County decision was not anti-AI. It was a disclosure and integrity standard, and agencies that want their AI-assisted reports to be usable in that jurisdiction need to meet that standard.

Brady and Giglio do not create new rules for AI. They apply the existing constitutional disclosure standard to AI-touched documents and reveal the consequences that have always applied to inaccurate police reports, now operating in a context where inaccuracy can be generated at scale and with fluent confidence.

The Chain of Custody Question

Chain of custody refers to the documented, unbroken record of who had possession of an item of evidence, when they had it, and what was done with it. In physical evidence, chain of custody is a well-established discipline: every transfer of a piece of physical evidence is logged, and any break in the chain can be used to challenge the admissibility or reliability of that evidence. The principle is that evidence that cannot be accounted for from collection to courtroom is evidence that may have been tampered with, contaminated, or confused with something else.

AI introduces a new chain-of-custody question in the context of documentary evidence. When an AI system processes body-camera footage and produces a draft report, a new document has been created by a process that is not a simple mechanical transcription. The model applied pattern-based generation, it made choices, and it produced text that reflects those choices as well as the underlying recording. If the draft contains any hallucinated content that survives verification and is adopted into the final report, the chain of custody for the facts in that report includes a generation step that may have introduced content that was not in the original footage.

The defense question in this context is not just "who held this evidence" but "how was this text produced?" If an officer testifies that the report is an accurate account of what the footage shows, and the defense can demonstrate that the report was produced by an AI system that can generate content not present in the footage, the defense has raised a chain-of-custody-style challenge to the accuracy of the documentary record.

This challenge is manageable. The way to manage it is the same way chain of custody for physical evidence is managed: documentation. An agency that documents that AI was used to produce the draft, that the officer conducted a footage-grounded verification pass, and that every statement in the adopted report was verified against the source material has a defensible chain of custody for that document. An agency that does not document any of this has left the question open, and open questions become cross-examination material.

Discovery, Deposition, and Cross-Examination

Understanding how police reports travel through the legal process helps clarify why verification and documentation standards need to be set before the report is submitted, not after a problem is discovered.

Discovery is the pre-trial process by which both prosecution and defense exchange evidence and information relevant to the case. In criminal cases, discovery includes the police report, the body-camera footage, the CAD entry, any other incident reports, and any other material the prosecution intends to use or is obligated to disclose under Brady and Giglio. When AI is used in report production, the defense may seek discovery of: the AI-generated draft, the tool used to generate it, the agency's policy for reviewing and verifying AI drafts, any logs showing what corrections the officer made, and documentation of the verification pass. Agencies that do not have documentation of these things cannot produce them, and courts have discretion to draw inferences from the absence of documentation that should have existed.

Depositions take the report further. In a deposition, the officer is under oath and answering questions about every statement in the report. "Did you observe this?" "Did the subject say this exact phrase?" "Is this detail in the footage?" These are not abstract questions. They are questions that require the officer to have personal, verified knowledge of every statement in the adopted report. An officer who adopted an AI draft after a cursory review, who cannot say with certainty whether each statement in the report is in the footage, is in a difficult position under oath. An officer who conducted a rigorous verification pass, documented it, and can point to the footage timestamp for every material claim is in a defensible position.

Cross-examination at trial takes this further still. Defense attorneys who understand how AI report-drafting works, and by 2026 many do, know the specific questions to ask. "Officer, did you draft this report yourself or did you use an AI tool?" "Did you verify this quote against the recording?" "Is this specific phrase present in the footage?" "What is your agency's policy for reviewing AI-generated drafts?" The officer who has done the work, verified the report, and followed the disclosure policy has honest, complete answers to all of these questions. The officer who has not is likely to produce answers that raise more questions, and raised questions in cross-examination are opportunities for the defense.

The Disclosure Obligation and the King County Lesson

The question of whether an officer's use of an AI tool must be disclosed to the prosecution, and through the prosecution to the defense, is one that jurisdictions are resolving in different ways. But the general direction of the legal and professional consensus, as of 2026, is clear: AI use in report production should be disclosed, and agencies that do not have a disclosure policy are behind the governance curve.

The King County, Washington example is instructive here. When the county's prosecuting attorney's office raised concerns about AI-generated police reports, the concern was not that AI was being used but that it was being used without documentation and without a clear standard for what the officer's verification had covered. The prosecutor's office needs to know, before it relies on a report in a charging decision or discloses the report as part of discovery, whether the statements in that report represent the officer's verified account of what the footage shows or whether they are AI-generated text that was adopted without rigorous verification. That distinction is relevant to every disclosure and charging decision the office makes.

An agency that documents AI use, requires verification passes, and discloses that documentation is an agency whose reports the prosecutor's office can work with. An agency that does not have those standards is an agency whose reports raise questions that prosecutors in some jurisdictions are now declining to work around, as King County demonstrated.

The Electronic Frontier Foundation (EFF) has separately raised transparency concerns about AI use in police documentation, from the community and civil-liberties perspective. The EFF's concern is that AI-assisted police documentation, if not disclosed and if not subject to adequate oversight, could be used to launder errors or biases into the legal record without accountability. These concerns are legitimate on their own terms, not just as compliance considerations. A community that cannot trust that police reports accurately reflect what happened is a community with reduced confidence in the legal process. An officer who understands why transparency matters in this context is an officer who can explain their AI use honestly, because they have done the work that makes honest explanation possible.

The Officer as Author, Not Editor

One of the most important conceptual shifts in this program is understanding the difference between being an editor of an AI draft and being the author of an adopted report. In an editing relationship, the editor reviews someone else's work, improves it, corrects obvious errors, and returns it. The underlying authorship still belongs to the original writer. In the legal context of a sworn police report, this framing does not apply. There is no legal concept of a report being co-authored by an officer and an AI system. The officer who submits the report is the author. Every statement in it is the officer's sworn account. The fact that the text was initially generated by an AI tool does not change this, and no court will accept "the AI wrote it" as an answer to "did you observe this fact?"

This is not a legal technicality. It is the right standard, for a reason that becomes clear when you think through what authorship of a sworn report actually means. The officer is the person who was there. The officer is the person who saw what happened, heard what was said, and observed the scene. The officer's report is supposed to be the officer's account of that experience, rendered accurately for the legal record. An AI tool can help the officer produce that account more efficiently, but the account is the officer's. The verification obligation is the process by which the officer confirms that the draft accurately represents their account. Completing that process is what makes the officer the actual author of the final report, not just a technical signatory.

The officer who reviews the draft against the footage, corrects the errors, removes the hallucinations, and adopts the corrected report as their sworn account is in a strong position: they have used AI to save time and they have met the authorship and verification standard. The officer who reviews the draft cursorily and adopts it with errors intact has not met the standard and is exposed to every consequence this lesson has described. The difference between these two outcomes is the verification discipline, and the verification discipline is a learnable, repeatable skill that this program teaches.

Key Takeaways

• A police report is not a memo or a working document. It is evidence that is disclosed to defense counsel, read at depositions, used as the basis for testimony, and tested at trial. This single fact reorders every rule about AI use in report production.
• Brady v. Maryland (1963) requires the prosecution to disclose any material exculpatory evidence to the defense. An AI hallucination in a police report, whether it adds a false incriminating detail or omits an exculpatory one, is a potential Brady problem. The footage, which is the ground truth, must match the report.
• Giglio v. United States (1972) requires disclosure of impeachment evidence, including evidence bearing on officer credibility. A documented pattern of an officer adopting AI drafts without adequate verification is Giglio material that must be disclosed to the defense.
• The King County, Washington prosecutor's guidance barring AI-written police reports without verification documentation illustrates the practical consequence of inadequate AI governance: prosecutors who cannot verify the provenance of report statements will decline to rely on those reports or will require additional documentation that agencies without adequate policies cannot provide.
• The EFF has raised legitimate transparency concerns about AI use in police documentation. Understanding these concerns is part of responsible AI use: an officer who can explain their AI use honestly and completely is an officer who has done the verification work that makes honest explanation possible.
• Chain of custody for documentary evidence now includes the question of how AI-generated text was produced and verified. Agencies that document their AI use, verification processes, and corrections maintain a defensible chain of custody. Agencies that do not leave that question open for cross-examination.
• Discovery requests in AI-touched cases can include the AI draft, the tool used, the agency's verification policy, and documentation of the officer's verification pass. Agencies without documentation cannot produce it, and courts can draw inferences from absent documentation.
• The officer who submits an AI-assisted report is the author of that report in every legal sense. There is no co-authorship with an AI tool. Every statement in the adopted report is the officer's sworn account. The verification process is the step that makes this true rather than nominal.