The assistant district attorney (DA) had not raised her voice. She did not need to. She slid a single printed page across the conference table toward Captain Luis Herrera and said, quietly, that her office would not be charging cases from his agency until the AI report issue was resolved. It was a Tuesday in October, three weeks after his department had begun rolling out Axon Draft One, the AI-assisted report-writing tool that drafts patrol narratives from body-worn camera (BWC) audio. Captain Herrera had not expected this meeting. He had not prepared for it. He left the DA's office with a legal problem he did not have a policy to solve.

The King County Objection, and What It Actually Means

In King County, Washington, the prosecutor's office formally barred AI-written police reports from cases it would charge. That decision was widely reported as a confrontation between law enforcement and prosecution, between new technology and old institutions. That framing misses the point. The King County decision was a governance failure, not a technology failure. The prosecutor's office lacked the documentation it needed to represent a report's reliability to a court. In the absence of that documentation, the prosecutor could not meet their own disclosure obligations under Brady v. Maryland (1963), which requires disclosure of exculpatory evidence, and Giglio v. United States (1972), which requires disclosure of impeachment material, including anything that could call into question the credibility of a government witness, including the officer who wrote the report.

"Brady" is shorthand for the constitutional obligation to disclose exculpatory evidence to the defense. "Giglio" extends that obligation to impeachment material. An AI-generated report that contains an unsupported detail, a gap-fill, a softened fact, or an invented quote is a Brady problem if it conceals exculpatory evidence, and a Giglio problem if it can be used to impeach the officer on the stand. A prosecutor who files a case without knowing whether the AI-drafted report contains any of those errors has taken on a risk they did not knowingly accept.

The King County objection, replicated in Captain Herrera's DA meeting and in dozens of similar conversations happening in 2026 across the country, is a request for three things: documentation of how the AI was used, documentation of how the officer reviewed and corrected the draft, and a disclosure standard that can accompany every case the agency sends to the prosecutor's office. It is not a rejection of the technology. It is a request for accountability. The agency that provides that documentation, in advance, as a matter of policy and workflow, meets the objection before it is raised.

The prosecutor's objection is not to AI-assisted reports. It is to AI-assisted reports that arrive without documentation of authorship, verification, and disclosure. Meet the objection in advance by building those three things into the workflow before the first case is charged.

What Prosecutors Need Before They Can File

A prosecutor receiving an AI-assisted report needs to be able to answer four questions before they can represent that report's contents to a court.

First: Who is the author? The law requires a human author for a sworn report. The AI is a drafting tool. The officer is the author. The prosecutor needs documentation that confirms this, not just an oral assurance. The documentation is the audit trail embedded in the report's metadata or appended as a disclosure statement: the officer's name, the date and time of review, the confirmation that corrections were made as needed, and the officer's adoption of the draft as their sworn account.

Second: How was the draft verified? The prosecutor needs to know that the officer checked the draft against the body-worn camera footage, the computer-aided dispatch (CAD) record, and their field notes. CAD is the timestamped dispatch record. The records management system (RMS) is the database in which the adopted report is stored. Knowing that a verification pass was completed, and that it followed a documented standard, allows the prosecutor to represent the report's reliability in good faith. A vague assurance that "officers review all AI drafts" does not satisfy that need. A documented verification standard, with a logged completion record, does.

Third: What corrections were made? If the AI draft contained an error and the officer corrected it, the prosecutor needs access to that correction history. Not because every correction is a Brady issue, but because a defense attorney who asks "was the original AI draft different from the submitted report?" deserves an accurate answer. If the answer is yes, and the correction history is logged, the prosecutor can disclose it and explain it. If the answer is yes, and no correction history exists, the prosecutor has a disclosure problem they cannot resolve.

Fourth: What tool was used, and what are its known limitations? A prosecutor who knows that Axon Draft One was used, that testing officers reported an 82 percent decrease in report-writing time, and that the tool drafts from audio which means it can produce gap-fills in sections with poor audio quality, is armed with the context they need to evaluate the report and to respond to defense challenges. The testing figure of 82 percent is a benchmark from officer testing, not a guarantee of accuracy; the prosecutor should understand that distinction. The fact that the tool exists on a Criminal Justice Information Services (CJIS) covered platform is relevant to any chain-of-custody question about the pre-adoption draft. This context should be documented and provided as part of the case package, not volunteered only when the defense attorney raises it.

Building the Relationship Before the Case Arrives

Captain Herrera's mistake was not deploying the AI tool. His mistake was deploying it without first sitting down with the prosecutor's office. The working relationship between an agency and its prosecutor's office on AI-assisted reports is not a legal formality. It is the practical foundation of every case the agency charges. That relationship should be built before the first AI-assisted report goes into a charging package, not after the first case is rejected.

The conversation should happen at the command-staff level, not the line-officer level. The chief or deputy chief meets with the lead prosecutor or the relevant division chief. The purpose of the meeting is not to sell the technology. It is to establish a shared protocol that both sides can operate under. That protocol should address the four questions above and produce a written document both offices can reference: what documentation accompanies AI-assisted reports, what the standard disclosure statement says, how correction histories are provided, and what the process is if the prosecutor needs additional documentation on a specific case.

The Joint Protocol Document

The joint protocol between the law enforcement agency and the prosecutor's office does not need to be long. It needs to answer five questions in writing, signed by both parties, before the first AI-assisted case is charged.

One: What is the disclosure statement? The specific text that appears in or with every AI-assisted report, naming the tool, the officer who reviewed it, the date of review, and the confirmation of adoption.

Two: What does the correction history contain? The format, the location in the case file, and the access process.

Three: What triggers a supplemental document? If a specific AI-assisted report presents a verification question the standard protocol does not resolve, what is the process for providing additional documentation to the prosecutor's office before filing?

Four: What training documentation can the prosecutor's office request? If a case goes to trial and the defense challenges the AI-drafting process, the prosecutor should be able to show that the officer was trained to the agency's verification standard. That training record should be accessible, documented, and available without requiring a subpoena.

Five: What is the escalation process? If a prosecutor reviewing a case file believes the AI-assisted report may contain an unresolved verification problem, who in the agency do they contact, and what is the expected turnaround time?

This protocol is a living document. It should be reviewed at the same cadence as the agency's AI use policy, at minimum annually, and updated whenever the tool changes significantly or a significant legal development occurs.

Preparing Officers for Cross-Examination on AI-Assisted Reports

The joint protocol protects the prosecutor in the charging phase. It does not protect the officer on the stand. That is a separate preparation problem, and it is one the agency must address in its training program.

A defense attorney who knows the report was AI-assisted will ask some version of the following questions in cross-examination: "Officer, did you write this report?" "Did a computer write part of this report?" "Were any changes made to the AI draft before you submitted it?" "How do you know that what the AI wrote was accurate?" "Did you personally verify every statement in this report against the body-camera footage?" These are not trick questions. They are legitimate challenges to the reliability of evidence. The officer who has completed the footage-grounded verification pass, documented it, and adopted the report with a clear understanding of their authorship role has straightforward answers to every one of them.

The answers sound like this: "I reviewed the AI-assisted draft, checked it against the body-worn camera footage, the dispatch record, and my field notes, made corrections where needed, and adopted it as my sworn account." That is a complete and defensible answer. It describes a specific process. It asserts personal knowledge. It does not disclaim authorship or responsibility.

The answer that destroys a case sounds like this: "I looked it over and it seemed right." Or: "The system generates the report automatically." Or: "I don't remember exactly which parts I changed." Each of those answers tells the jury that the officer did not understand their role in the process, did not exercise the judgment the law requires, and may not be a reliable narrator of the events in the report. The training program must be specific enough, and the policy clear enough, that the first set of answers is what officers give, not the second.

Anatomy of a Suppression Motion Based on AI Reports

Command staff and legal advisors should understand the likely form of a suppression motion targeting an AI-assisted report, because understanding the attack is the first step in building the defense. The motion will argue one or more of the following:

Brady violation: The report contained a gap-fill, softened fact, or invented detail that the defense characterizes as exculpatory, and that detail was not disclosed as a product of AI generation.

Giglio violation: The officer's use of an undisclosed AI tool is itself impeachment material, because it goes to the reliability of the officer's account, and that information was not provided to the defense.

Foundation failure: The report does not establish personal knowledge as required for admissibility, because the officer cannot testify that they personally wrote every statement in it.

Chain of custody: The pre-adoption AI draft, which may have been stored on a vendor's cloud server before adoption, was not subject to the same evidence-preservation obligations as an adopted report, creating a gap in the chain of custody for the document.

Each of these arguments is defeatable with the right documentation. The Brady and Giglio arguments are defeated by the disclosure statement and the correction history: the agency disclosed the AI's role, the officer corrected errors, and both are documented. The foundation argument is defeated by the officer's deposition testimony about the verification process and their adoption of the document as their sworn account. The chain-of-custody argument is defeated by the vendor contract's data retention provisions, which should specify how pre-adoption drafts are handled and retained.

The agency that has built this documentation into its workflow is not surprised by any of these arguments. The agency that has not built it in is spending its litigation budget responding to each one of them, case by case, with no consistent answer.

When the Prosecutor Still Objects: Escalation and Adaptation

Captain Herrera went back to the DA's office two weeks after the first meeting. He brought a draft joint protocol, a copy of the agency's use policy, and a sample disclosure statement. He also brought the agency's legal advisor. The conversation was still difficult, but it was different. The prosecutor's office had a document to respond to rather than a gap to refuse.

There will be prosecutors who remain skeptical of AI-assisted reports even after a joint protocol is in place. That skepticism is not unreasonable. The technology is new. The case law is thin. A prosecutor who has seen a gap-fill turn into a suppression motion is going to apply a higher standard for the next AI-assisted case, and that is their right. The agency's obligation is not to overcome skepticism by argument. It is to overcome it by evidence: the audit trail, the correction history, the verification logs, the training records.

If a specific prosecutor refuses to file AI-assisted cases under any circumstances, the agency should escalate to the elected DA or the office's legal director. Not to override the line prosecutor's judgment, but to get clarity on the office's actual policy. An individual line prosecutor's discomfort with AI tools is a different question from a formal office-wide prohibition. Understanding which one the agency is dealing with determines the appropriate response.

If the prosecutor's office maintains a blanket prohibition, the agency must decide whether the efficiency gain from AI-assisted reports is worth the friction of working around that prohibition for every case that reaches the DA. That is a strategic decision for command staff and the city manager, not a policy question the patrol commander resolves unilaterally. The use policy, the joint protocol, and the training program are the levers. The command staff's job is to make sure those levers are actually deployed, not just documented.

One structural hedge is worth naming: the agency should build its AI-assisted report workflow so that the human officer's role is visible and documentable independent of the AI tool. If the vendor changes its platform, if the contract ends, or if a specific tool becomes legally contested, the officer's verification pass, their correction history, and their adoption record should be portable: stored in the agency's RMS, not exclusively in the vendor's cloud. CJIS obligations stay with the agency; so do evidence obligations. Vendor independence in the audit trail is not just good governance. It is what allows the agency to answer discovery demands without going back to the vendor for every document.

Key Takeaways

• The King County-style prosecutor objection to AI reports is not a rejection of technology. It is a request for documentation of authorship, verification, and disclosure. Meeting that request in advance, before the first case is charged, prevents the objection from becoming a moratorium.
• Brady v. Maryland and Giglio v. United States make AI-assisted report writing a constitutional disclosure question. If the method of generating a report is material to the defense's ability to challenge it, that method must be disclosed. The prosecutor needs this documentation to file.
• The working relationship between the agency and the prosecutor's office on AI reports must be built at the command-staff level, before the first AI-assisted case is charged, and documented in a joint protocol that both offices sign.
• The joint protocol should answer five questions in writing: what is the disclosure statement, what does the correction history contain, what triggers a supplemental document, what training documentation is available, and what is the escalation process.
• Officers must be trained to answer cross-examination questions about AI-assisted reports with specific, process-grounded answers, not general assurances. "I reviewed the AI-assisted draft, checked it against the footage and the CAD record, made corrections where needed, and adopted it as my sworn account" is the answer that survives cross-examination.
• The four likely suppression-motion arguments targeting AI-assisted reports (Brady violation, Giglio violation, foundation failure, and chain-of-custody gap) are each defeated by specific documentation: the disclosure statement, the correction history, the deposition testimony about the verification process, and the vendor contract's data retention provisions.
• If a specific prosecutor maintains a blanket objection, escalate to the elected DA or the office's legal director for clarity on the actual office policy. An individual line prosecutor's discomfort is a different problem from a formal prohibition, and deserves a different response.
• The audit trail, the correction history, and the verification logs should be stored in the agency's RMS, not exclusively in the vendor's cloud. Vendor independence in the audit trail is what allows the agency to answer discovery demands without relying on the vendor for every document.