Chief Angela Reyes had been running her city's police department for six years when the vendor pitch landed on her desk: a bundled body-worn camera (BWC, the recording device each officer clips to their uniform), cloud storage, and AI-assisted report-drafting package, seven hundred officers, roughly $45 million over ten years. She had three people who understood the technology well enough to have an opinion. None of them had it in their job description. One was a patrol sergeant who'd read about Axon Draft One on his own time. One was a records clerk who'd noticed the redaction tool behaving strangely on domestic-violence footage. The third was the department's IT manager, who knew the Criminal Justice Information Services (CJIS, the FBI's security policy that governs how criminal-justice data must be stored, transmitted, and protected) obligations but had never worked a criminal case. Reyes signed a letter of intent. Six months later, a defense attorney filed a motion to suppress every AI-assisted report the department had produced. The motion cited Brady v. Maryland, which requires the prosecution to disclose any evidence favorable to the defendant, including evidence about how that evidence was produced. The department had no disclosure policy, no AI use log, and no one whose job it was to own any of those things.

The Org Chart Problem: No One's Job

The scene above is not hypothetical in any meaningful sense. It is a composite of conversations happening in police departments across the country as of 2026, as agencies absorb AI tools faster than they are building the governance structures to operate them safely. The pattern is consistent: the tool arrives before the role, the capability before the accountability, the contract before the policy.

This lesson is about fixing that pattern at the organizational level. Not the policy document, which is covered elsewhere in this program. Not the technical configuration, which the vendor handles in part. The question here is simpler and more human: whose job is this?

In a department that is using AI tools responsibly, three roles must exist. They can be full-time positions, part-time designations, or components of existing jobs, depending on agency size. But they must exist as named, accountable functions, not distributed assumptions. The three roles are the Agency AI Lead, the Disclosure Coordinator, and the Review Cadre.

Together they form what practitioners are beginning to call the AI governance triangle: the Lead provides strategic direction and accountability upward to command and city leadership; the Disclosure Coordinator manages the legal obligation that AI-assisted materials are properly documented and disclosed to prosecutors and defense counsel; and the Review Cadre provides the operational quality check that AI outputs are verified before they become sworn accounts or official records. Without all three, the triangle is missing a corner. Without any of them, you do not have governance. You have hope.

When no one's job description includes AI oversight, AI oversight does not happen. Distributed responsibility is the organizational equivalent of no responsibility.

The Agency AI Lead: What the Role Actually Requires

The Agency AI Lead is the person whose professional responsibility includes the agency's AI program. They are accountable for whether the program operates within policy, within the law, and within the community's trust. They report to command, speak to city council when asked, and meet with the prosecutor's office regularly enough that neither side is surprised by what the other is doing.

The Specific Knowledge the AI Lead Must Carry

This is not a technology role in the narrow sense. The Agency AI Lead does not need to be able to train a machine-learning model or audit source code. They need to be able to answer four questions correctly and under pressure: What AI tools are we using? What is each tool authorized to do? How is each tool's output reviewed before it becomes official? Who is accountable for each tool's failures?

The AI Lead must understand, at minimum, how Axon Draft One and comparable tools produce a narrative from BWC audio: the transcription-to-draft pipeline, the hallucination failure modes (specifically the gap-fill detail, the softened fact, and the invented quote, each of which becomes a Brady problem if it reaches a sworn report), and the disclosure obligations that attach when AI touches a file that may be used in prosecution.

They must understand the CJIS Security Policy and its implications for vendor cloud storage. CJIS obligations stay with the agency. A vendor's commitment to CJIS compliance does not transfer the obligation; it only shapes how the agency meets it. The AI Lead needs to be the person who can tell a city council member, in plain language, what data the vendor holds, under what security standards, and what the agency's audit rights are. The data governance question and the evidentiary question are the same question. When criminal-justice data leaves the agency's direct control, the AI Lead needs to know where it goes and under what terms.

They must be familiar with Brady v. Maryland and Giglio v. United States. Brady, decided in 1963, requires prosecutors to disclose any evidence favorable to the defendant. Giglio, decided in 1972, extends that obligation to impeachment evidence about witnesses, including the officers who wrote the reports. If AI was used to draft a report and the draft contained a significant error that was caught and corrected, that correction history may be Brady or Giglio material. If an agency is using AI tools and has not worked through that question with the prosecutor's office, the AI Lead's first meeting with the prosecutor's office should be about exactly that.

They must know the King County precedent. In 2023, the King County, Washington prosecutor's office barred AI-written police reports, citing concerns about accuracy and the inability to assess how the draft was verified. That decision is a governance marker: it is what happens when disclosure and verification are not built in. The AI Lead's job is to build a program the prosecutor's office does not need to bar.

The AI Lead as the Agency's Public Face on AI

The AI Lead will be asked to explain the program to audiences who are skeptical. Community members who have heard about AI policing tools and have legitimate concerns. Defense attorneys who want to understand the disclosure scope. Oversight board members who need a straight account of what the technology does and does not do. Reporters covering a suppression motion who call and ask what the department's policy is.

The Electronic Frontier Foundation (EFF), a civil-liberties organization that has consistently monitored AI deployment in law enforcement, has raised concerns about transparency in AI-assisted police reports: specifically, that agencies are deploying these tools without adequate public disclosure about how they are used, how errors are caught, and how the tools' outputs are documented in discovery. The EFF's concerns are the same concerns the AI Lead needs to be able to answer, directly and without defensiveness. An AI Lead who can sit across from an EFF representative or a city council oversight committee and say "here is our use policy, here is our verification standard, here is our disclosure protocol, and here is the audit log for the past six months" is doing the job. An AI Lead who says "I'll have to look into that" when the motion is already filed is not.

Who Should Hold This Role

In a large department, the Agency AI Lead may be a dedicated position, possibly a lieutenant or captain with both operational experience and comfort with technology policy. In a mid-size department, it may be an assignment given to a ranking officer or civilian administrator who completes additional training and carries the designation explicitly in their performance evaluation. In a small department, it may be the chief themselves, or a deputy assigned the function. What it cannot be is no one.

The role should be reflected in a formal job description or designation letter, not just a verbal understanding. The designation should specify what decisions the AI Lead has authority to make unilaterally, what decisions require command approval, and what reporting cadence the AI Lead has to both command and external oversight. A role without those specifics is a title, not a function.

Salary and promotional credit matter here, not as a budget line but as a signal. If the AI Lead designation comes with no career benefit and no protected time, the person holding it will treat it as a side duty, because that is what the organization is telling them it is. If catching a Brady problem in AI-generated reports before it reaches trial is worth doing, it is worth incentivizing. The agency that puts the AI Lead on a performance evaluation plan with measurable outcomes is the agency that actually gets AI governance. The one that gives the role to whoever raised their hand at a staff meeting is likely to learn this lesson the hard way.

The Disclosure Coordinator: The Legal Linchpin

The Disclosure Coordinator occupies the seam between the agency's AI program and the criminal-justice system's disclosure obligations. If the AI Lead is the strategic owner of the program, the Disclosure Coordinator is the operational owner of the question that most directly affects case outcomes: when AI touches a file, what must be disclosed, to whom, and by when?

Why Disclosure Is a Separate Function

Before AI-assisted report writing became standard, disclosure in the context of a police report was relatively straightforward: the report itself was disclosed to the defense as part of discovery. What changed with AI is that the report now has a history. The AI draft is one version. The verified, corrected, adopted report is another. Any significant discrepancy between them is potentially Brady material. The AI tool's configuration, its training data's characteristics, its known failure modes are potentially Giglio material if a challenge to the report's accuracy would impeach the officer who signed it.

The records management system (RMS, the agency's case documentation platform) may or may not preserve the AI draft alongside the adopted report. If it does not, and if a defense attorney demands the original AI output in discovery, the agency may not have it. That is a discovery failure with legal consequences. If the RMS does preserve it but no one has told the prosecutor's office that the tool produces a draft that is distinct from the adopted report, the prosecutor may not know to include that draft in the disclosure package. That is also a discovery failure.

The Disclosure Coordinator's job is to make sure neither of these failures happens. They design and maintain the disclosure protocol for AI-assisted materials. They brief the prosecutor's office on what the agency is disclosing and why. They respond to defense requests for AI-related materials with a clear, auditable account of what materials exist. And they work with the AI Lead to flag when the existing disclosure framework needs to be updated because the tools have changed.

The Disclosure Coordinator's Workflow

The Disclosure Coordinator's day-to-day work involves four recurring functions. First, they maintain the AI use log: a record of which reports were AI-assisted, which tool was used, what the disclosure flag in the RMS says, and when the log was last reviewed. In a department using AI for report drafting across hundreds of officers, this log is not a spreadsheet maintained by hand. It is a structured data export from the evidence and report platform, reviewed on a defined cadence by the Disclosure Coordinator against a checklist.

Second, they manage the disclosure language template. Every AI-assisted report should include a standard disclosure notation (something on the order of: "The initial narrative for this report was drafted with the assistance of Axon Draft One using the body-camera recording and reviewed, corrected, and adopted by the reporting officer"). The Disclosure Coordinator maintains that template, updates it when the tool or process changes, and ensures it is actually being used by reviewing a sample of reports on a regular schedule. The computer-aided dispatch (CAD) system and the RMS should both reflect AI assistance for any incident where a draft was generated, not just the ones where the officer remembers to flag it.

Third, they maintain a running brief for the prosecutor's office. This is not an adversarial document. It is a cooperative one. The prosecutor needs to know what the agency's AI program looks like, what the verification standard is, what the disclosure package for an AI-assisted case includes, and what to tell the defense if the defense asks. A prosecutor who knows the answers to those questions is not surprised by a defense motion. A prosecutor who does not know is potentially blindsided in a Brady hearing, and the relationship between the agency and the prosecutor's office suffers for it.

Fourth, the Disclosure Coordinator maintains an incident log for AI-related errors. When an officer catches a significant gap-fill during the verification pass, when a discrepancy between the AI draft and the footage is documented, or when a case is challenged on AI-related grounds, that event goes into the incident log. The log does not exist to assign blame. It exists to provide the data needed to answer the question: is the AI tool performing within acceptable error bounds, and is the verification process catching errors before they reach sworn reports? Without that data, the answer to that question is a guess. With it, the answer is an evidence-based operational assessment the AI Lead can bring to command and council.

Sizing the Disclosure Coordinator Function

In a large department, the Disclosure Coordinator may be a full-time position, most naturally seated in or adjacent to the legal affairs unit. In a mid-size department, it may be a 25 to 50 percent designation for a records supervisor or paralegal with criminal-justice disclosure experience. In a small department, it may be a function the chief or a senior records clerk performs with structured support from a template and a quarterly review cadence.

What the function cannot be is informal. "We disclose what we're supposed to" is not a policy. It is an aspiration, and it is not enough when a defense attorney files a Brady motion and asks to see the agency's AI disclosure documentation. The Disclosure Coordinator role exists to turn the aspiration into a documented, auditable practice.

Brady and Giglio obligations did not change when AI entered the report-writing workflow. What changed is the number of things that may need to be disclosed and the discipline required to track them all.

The Review Cadre: Quality at Scale

The Review Cadre is the group of officers, supervisors, and subject-matter specialists who provide quality assurance on AI-assisted outputs across the agency's operations. The individual officer's verification pass catches errors in a single report. The Review Cadre's function is to catch patterns: systematic errors, failure modes that appear repeatedly across a class of incidents, verification gaps that suggest the training program is not working, and tool behaviors that need to be escalated to the vendor or the AI Lead.

What the Review Cadre Does

The Review Cadre is not an internal affairs function. It is a quality function. The distinction matters for culture, and it matters for recruitment to the cadre. Officers and supervisors who join the cadre are doing quality assurance on a technical process, not investigating colleagues. The goal is to find errors the process should have caught and improve the process, not to sanction the individuals who did not catch them. (That is a separate function, and a separate conversation.)

The cadre reviews a defined sample of AI-assisted reports on a regular cadence, typically monthly, checking for three things. First, are the reports correctly flagged as AI-assisted in the RMS, and does the disclosure notation appear? Second, are there factual claims in the adopted report that do not appear to be grounded in the footage or the CAD entry, suggesting a gap-fill that the officer's verification pass missed? Third, are there patterns across the sample: a specific call type where gap-fills appear most often, a specific tool configuration that produces more errors, a specific unit or shift where the verification pass may need reinforcement?

The cadre also reviews dispatch AI outputs if the agency is using AI in computer-aided dispatch. The stakes in dispatch are, in some respects, higher than in report writing, because a misclassified call is not an evidentiary problem for a future case. It is a response problem for a current emergency. A call classified as a property disturbance that is actually a domestic violence incident in progress is a life-safety failure. The cadre reviews sampled CAD entries for AI-suggested classifications against the call's actual audio, looking for patterns in misclassification, particularly for call types where the consequences of error are most severe.

Who Sits on the Review Cadre

The cadre should be cross-functional: representation from patrol, from dispatch if the agency is using AI in that context, from records, and from the legal affairs or prosecutor-liaison function. A cadre composed only of patrol officers will miss systematic errors in redaction and records release. A cadre composed only of records staff will miss the use-of-force gap-fills that a patrol-experienced reviewer would recognize immediately.

The cadre should include at least one person with enough understanding of the AI tool's output characteristics to distinguish a normal output from an anomalous one. This does not require a data scientist. It requires someone who has seen enough AI drafts to know, when they read a sentence, whether it reads like something the footage would support or something the model assembled from training-data patterns. That kind of judgment develops with experience, and it should be cultivated deliberately within the cadre.

Service on the cadre should rotate, with enough overlap that institutional knowledge about error patterns is preserved across the rotation. A cadre member who has seen three months of reports and knows that the tool consistently mishandles overlapping voices in high-stress arrest situations is carrying valuable knowledge that should not walk out the door when they rotate off. Document it. Put it in the cadre's running error log, which feeds the incident log the Disclosure Coordinator maintains and the performance data the AI Lead brings to command.

The Review Cadre as the Agency's Early Warning System

The single most valuable function the Review Cadre performs is the one nobody wants to need: early warning. When the tool starts producing a class of error that was not there before, when verification quality drops across a unit, when a new call type is producing systematically bad outputs, the cadre is the mechanism that surfaces that signal before it becomes a suppression motion, a Brady hearing, or a community-trust crisis.

In 2026, agencies are signing multi-year, multi-million-dollar contracts (the bundled camera-drone-cloud-AI packages that can run $45 million or more over ten years) with vendors whose tools are still evolving. The tool that performed within acceptable error bounds at contract signing may not perform the same way after a major model update. The Review Cadre's monthly sample review is the mechanism that would catch that drift. Without it, the first signal of a problem may be a challenge in court.

The cadre should report to the AI Lead on a quarterly basis, with a written summary of findings: sample size reviewed, error rate observed, error patterns identified, and recommended actions. Those reports become part of the agency's AI program documentation. They demonstrate that oversight is real, ongoing, and responsive to actual data. They also provide the evidence base for the AI Lead's conversations with command and council about whether the program is working.

The Review Cadre's job is not to catch bad officers. It is to catch bad outputs before they become bad outcomes. The distinction keeps the function operating as quality assurance rather than surveillance.

How the Three Roles Work Together

The governance triangle functions because the three roles share information and have defined handoffs. The Review Cadre's error log feeds the Disclosure Coordinator's incident log and the AI Lead's performance data. The Disclosure Coordinator's Brady and Giglio flag process is informed by what the cadre is finding in review. The AI Lead's conversations with command and the prosecutor's office are grounded in data from both the cadre and the Disclosure Coordinator, not in impressions and hopes.

Consider what that loop looks like in practice. The Review Cadre reviews a month of AI-assisted reports and finds that in calls involving vehicle stops, the AI tool is systematically producing CAD (computer-aided dispatch, the system that tracks calls and units) entries that omit the number of occupants in the vehicle when that information was provided by the officer on camera. The cadre documents this pattern, marks the specific reports, and reports the finding to the AI Lead.

The AI Lead takes two immediate actions. First, they notify the Disclosure Coordinator that a class of reports in that time window may have incomplete CAD entries, and that the prosecutor's office should be briefed on the finding and invited to review the affected cases. Second, they contact the vendor with a documented description of the failure mode, requesting a timeline for a fix and an assessment of how many reports in the agency's system may be affected.

The Disclosure Coordinator briefs the prosecutor's office, provides the affected report numbers and the description of the gap, and works with the prosecutor to assess whether any of the affected cases have discovery implications. In cases where the missing occupant information was not relevant to the charge, the disclosure is brief and the impact is minimal. In cases where the number of occupants in the vehicle was material to the case (a search-and-seizure question, a witness identification question), the Disclosure Coordinator works with the prosecutor and defense to ensure the gap is fully documented in the case file.

All of this happens without a suppression motion, without a Brady hearing, and without a community-trust crisis, because the governance triangle caught the problem at the quality-assurance stage rather than at the trial stage. That is the value the three roles create. It is not visible when it works. It is very visible when it does not.

Standing Up the Roles: A Practical Sequence

The question for a chief, a deputy chief, or an HR lead reading this lesson is not whether to create these roles. The question is how to stand them up in a way that is durable, not just compliant with a current political moment. The following sequence works at most department sizes, with adjustments for agency scale.

Start with the Disclosure Coordinator

If the department is already using AI tools in report writing or dispatch, the most urgent role to formalize is the Disclosure Coordinator. Brady and Giglio obligations do not wait for organizational planning. If the agency is producing AI-assisted reports and has not established a disclosure protocol, that gap is already creating exposure in cases currently in the system. The Disclosure Coordinator function can be stood up quickly, with an existing records supervisor or legal affairs staff member given the designation, a template to work from, and a first meeting with the prosecutor's office scheduled within thirty days.

The disclosure template does not need to be elaborate. It needs to be consistent. Every AI-assisted report uses the same disclosure language. The RMS reflects the AI-assistance flag. The prosecutor's office knows what the disclosure package for an AI-assisted case includes. That minimum viable disclosure function can be operating within four to six weeks, before the AI Lead is formally designated and the Review Cadre is staffed.

Designate the AI Lead with Explicit Authority

The AI Lead designation should come with an explicit scope of authority in writing. What decisions can the AI Lead make alone? Revising the disclosure template. Escalating a vendor concern to procurement. Pausing AI-assisted drafting in a specific unit pending a quality review. What decisions require command approval? Changing the agency's AI tool configuration. Entering into a new vendor agreement. Modifying the verification standard. What requires city council notification? Any change that affects the public disclosure posture. Any significant incident with AI-related causes. Any budget decision above a defined threshold.

The written scope matters because it tells the organization that the AI Lead role has actual authority, not just advisory status. A role with advisory status and no decision authority is a role that will be ignored when it is inconvenient. The AI Lead needs to be able to act when the Review Cadre surfaces a problem, not draft a recommendation memo and wait for command to decide whether to act.

Build the Review Cadre from Existing Expertise

The Review Cadre does not require new hires. It requires structured time and a defined mandate. Identify three to five people across patrol, dispatch, and records who have demonstrated engagement with the AI tools and whose operational experience gives them the judgment to assess whether a report's claims are well-grounded. Give them protected time for the monthly review function. Define the review protocol in writing. Connect them to the AI Lead and the Disclosure Coordinator with a clear reporting chain.

The cadre's first task should be a retrospective review of a sample of AI-assisted reports from the past six months. Not to find fault, but to establish a baseline: what is the current error rate? What types of errors are most common? What is the current disclosure compliance rate? That baseline is the starting point for the cadre's ongoing quality function, and it is also the first data the AI Lead can use to report to command on the program's current state.

A note on scale: the Review Cadre in a 50-officer department looks different from the one in a 1,000-officer department. In a small department, the cadre may be two supervisors who split the review function across a month, using a checklist, and reporting findings to the chief. In a large department, it may be a structured cross-functional team with a dedicated meeting cadence and a formal reporting template. The principles are the same. The infrastructure scales with agency size. What does not scale down is the function itself. Every agency using AI tools needs someone reviewing the outputs on a systematic basis.

Key Takeaways

• Three named, accountable roles form the organizational foundation of a responsible AI program: the Agency AI Lead, the Disclosure Coordinator, and the Review Cadre. Without all three, governance is an aspiration, not a practice.
• The Agency AI Lead is accountable for the entire program to command, city leadership, and the prosecutor's office. They must understand CJIS obligations, Brady v. Maryland and Giglio v. United States implications, the King County precedent, and the AI tool's specific failure modes well enough to explain all of them under pressure.
• The Disclosure Coordinator owns the legal linchpin: documenting AI assistance, maintaining the disclosure protocol, briefing the prosecutor's office, and tracking AI-related errors in an incident log. Brady and Giglio do not pause while the agency figures out its disclosure process.
• The Review Cadre provides systematic quality assurance across a sample of AI-assisted outputs, identifying error patterns that individual officer verification would not surface. They are the early warning system the agency needs before a suppression motion becomes the first signal of a systemic problem.
• The three roles must share information through defined handoffs: the cadre's error findings feed the Disclosure Coordinator's incident log and the AI Lead's performance data. The loop only closes when all three are talking to each other on a defined schedule.
• Formalization matters. Job descriptions, written authority scopes, designated time, and connection to performance evaluations are what turn a named role into an effective function. A role that exists only in conversation is not a role the organization will rely on when it counts.
• Start with the Disclosure Coordinator if AI tools are already deployed, because Brady and Giglio exposure begins the moment the first AI-assisted report is produced without a disclosure protocol. The other roles follow, but that one cannot wait.
• The governance triangle is the mechanism that allows an agency to say, with documentation, to a prosecutor, a defense attorney, an oversight board, or a community: here is what we are using, here is how it is reviewed, here is who is responsible, and here is the evidence that the oversight is real.