The deputy chief walked out of the city council chambers at 9:47 PM and immediately called the chief. "They killed it," she said. The council had voted to pause the agency's AI-assisted report-writing rollout after a council member read a single EFF (Electronic Frontier Foundation, the civil liberties organization that has raised transparency concerns about AI police reports) article aloud and asked a question the deputy chief could not answer: "Do you disclose to defendants when AI wrote their arrest report?" The question was fair. The answer required more than thirty seconds. And in the silence between the question and the stammered response, the council read a program in trouble.

Why Alignment Is Not Agreement

The goal of working with three audiences, command staff, city council, and the community, is not to produce agreement. Agreement on a complex, evolving technology program in a politically sensitive field is too high a bar and often the wrong bar. The goal is alignment: a shared factual understanding of what the agency is doing, why, what safeguards exist, and how problems are caught and corrected. Alignment allows the council member to answer their constituent's question with the same factual account the chief would give. Alignment allows a community advocate to understand what they are looking at when they file a public records request about the AI program. Alignment allows the command staff to enforce the verification standard because they understand why it is the standard, not just that it is.

The deputy chief's failure was not a communications failure. It was an alignment failure that showed up as a communications failure. The program had not built the factual foundation with the council before the council was asked to weigh in on it. The EFF question, entirely predictable, was not in the deck. The Brady v. Maryland (the 1963 Supreme Court case requiring disclosure of exculpatory evidence to the defense) and Giglio v. United States (the 1972 case requiring disclosure of impeachment evidence including information affecting officer credibility) implications of AI in the evidence chain were not in the deck. The King County, Washington, prosecutor's decision to bar AI-written police reports was not in the deck. The result was a pause that cost the agency four months and required a full restart of the stakeholder process.

Alignment built before the program is under scrutiny costs a fraction of alignment rebuilt after trust has been damaged. The math is not complicated. An executive who spends twenty hours building the factual foundation with three audiences before rollout does not spend four months rebuilding it after a council vote that went wrong.

The same factual account, told at different depths for different audiences, is not spin. It is the definition of honest governance.

The Command Staff Audience

Command staff, meaning the deputy chiefs, division commanders, lieutenants, and supervisory sergeants who translate executive direction into daily operations, is the audience with the highest operational stakes in the AI program. A patrol lieutenant whose officers are using Axon Draft One (the AI product that drafts police narratives from body-worn camera, or BWC, audio and has shown an 82 percent reduction in report-writing time in officer testing) needs to understand the program deeply enough to supervise it, catch failures, and answer officers' questions with authority. A deputy chief who cannot explain the Brady implications of an AI-assisted report to the chief cannot explain them in a suppression hearing either.

The command staff conversation has a specific content requirement. It must cover the operating model: what tools are authorized, for what purposes, under what review conditions. It must cover the verification standard: what does the footage-grounded pass require, why is it not optional, and what happens when an officer skips it. It must cover the error log: what the pilot's gap-fill data showed, what categories of error are most common, and what correction procedures exist. It must cover the disclosure standard: exactly what must be documented and provided to the prosecutor's office when AI touched a case, and the format that the prosecutor's office has accepted. And it must cover the accountability structure: who is the agency AI lead, who is the disclosure coordinator, and who sits on the review cadre.

Command staff alignment is not a single briefing. It is a sustained relationship that includes regular reporting on the program's performance metrics. Officers spending 30 to 40 percent of every shift on paperwork before AI was deployed is a figure supervisors understand viscerally; the fact that AI-assisted reporting returned a measurable fraction of that time to street patrol, investigations, and community contact is a concrete operational benefit. That benefit is real, and command staff needs to understand it and own it. The risk is equally real, and command staff needs to understand that with the same depth. A lieutenant who can explain both to a skeptical officer is the operational backbone of the program.

The Deposition-Ready Commander

The highest test of command staff alignment is this: can the commander answer the deposition question? "Commander, your agency uses AI to write police reports. What exactly did the AI do in this case, how did the officer verify it, and what was disclosed to the defense?" A commander who has not been aligned on the program at depth cannot answer that question. The answer requires knowing the specific tool, the specific verification requirement, the specific disclosure format, and the specific audit trail. That is not generic AI knowledge. That is program-specific operational knowledge that must be built deliberately through alignment.

The answer the commander is aiming for is something like: "In this agency, AI-assisted reports are produced by [tool], which generates a draft from BWC audio. The officer is required to verify every factual claim against the footage before adopting the draft as their sworn account, and that verification is logged. When AI assistance was used, that is documented in the disclosure package provided to the prosecutor's office, in the format the prosecutor's office approved, along with the officer's verification log. The audit trail for this specific report is part of the production in discovery." That answer takes twenty seconds and closes the inquiry. It takes months of alignment to be able to give it with authority.

The City Council Audience

City council members, county commissioners, and elected governing boards occupy a different position in the alignment triangle. They are stewards of public resources and community trust, not operations experts. They are accountable to constituents who have read the same EFF articles, who have heard about the King County prosecutor's concerns, and who may have strong feelings about AI in criminal justice regardless of the specifics. The council member who asks about Brady disclosure when they hear the word "AI" in the same sentence as "police report" is not being obstructionist. They are doing exactly what their constituents elected them to do.

The council presentation has a specific architecture. It leads with the problem the program solves: officers are spending 30 to 40 percent of every shift on paperwork. That time is not discretionary. It is time not spent on patrol, on investigations, on community contact. The AI-assisted workflow recovers some of that time at a validated scale. That is the efficiency case, stated in operational terms the council can understand and defend to their constituents.

The presentation then addresses the safeguards immediately, not as a defensive response to anticipated objections but as an integral part of the program architecture. These are: the verification requirement (every AI draft is checked by the officer against the footage before it becomes a sworn report), the disclosure requirement (the prosecutor's office has been briefed, the disclosure format has been reviewed, the agency meets the obligation that the King County prosecutor's action established as the standard), the audit trail (every AI-touched report has a logged chain of review), and the accountability structure (named roles with specific responsibilities).

Then, and only then, the performance data. The gap-fill detection rate from the pilot, the disclosure compliance rate, the error-correction log. The council needs to see that the program is measuring its integrity, not just its speed. A presentation that leads with efficiency numbers and buries safeguards is a presentation that will be challenged; a presentation that treats safeguards as equal in importance to efficiency is a presentation that can survive scrutiny.

The Council Member Who Walks In Hostile

In most jurisdictions, at least one council member will walk into the AI program presentation with skepticism or outright opposition. This is not a problem to eliminate. It is a constituency to align. The council member who walks in believing that AI police reports are inherently biased, or that they undermine defendants' rights, or that they are a surveillance tool in disguise, is carrying legitimate concerns rooted in legitimate evidence, not all of which is wrong.

The most effective approach is to address the specific concerns with specific information, not to argue that the concerns are misplaced. "You are right that the King County prosecutor barred AI-written reports. Here is what that bar was based on, and here is specifically how our program meets the standard that prompted the bar." That answer respects the concern, demonstrates program knowledge, and moves the conversation from ideology to specifics. An executive who argues that AI concerns are overblown has lost the council member. An executive who says "here is the legitimate version of your concern and here is our specific response" has a chance.

The CJIS (Criminal Justice Information Services) Security Policy obligations, data handling standards that govern how criminal justice information is stored and transmitted, are also a council-level issue when the AI program involves cloud storage of criminal justice information. Council members whose constituents are concerned about data security need to know: what CJIS obligations apply, how the agency meets them, and what happens if the vendor has a breach. The answer to the last question (the agency remains accountable regardless of vendor fault) is not reassuring, but it is honest. An honest, uncomfortable answer that the agency has a plan for is more credible than a reassuring answer that proves false later.

The Community Audience

The community audience is the largest, the most diverse, and in many jurisdictions the most important for the long-term legitimacy of the program. Community trust in law enforcement is not a soft metric. It affects cooperation with investigations, willingness to report crime, participation in community policing programs, and the political support that sustains the agency's budget and operational authority. An AI program that the community does not trust is a liability even if it is operationally sound.

Community engagement on an AI program is not a town hall event. It is a sustained process that begins before deployment and continues through the program's life. The content of that process changes over time, but the structure is consistent: the community receives factual information about what the program does and does not do; the community has a meaningful mechanism to raise concerns; those concerns are acknowledged, responded to, and where appropriate incorporated into program adjustments; and the results are communicated back to the community in plain language.

The EFF's specific concern is opacity. The community cannot see what the AI is doing, cannot assess whether the verification is real, and cannot evaluate whether the disclosure obligations are being met. The agency that turns that concern into a program feature, by making the program's operations genuinely transparent, has converted a civil liberties adversary into a potential partner. This does not mean releasing sensitive enforcement data. It means: here is the policy, here is what the AI does, here is what the officer checks before the report is adopted, here is what is disclosed to the defense, here is how you can request information about AI use in a specific case, and here is how to raise a concern.

Community Oversight Bodies as Partners

Many jurisdictions have civilian oversight boards, police commissions, or inspector general offices that review agency conduct. These bodies are not adversaries in the context of an AI program; they are the institutional mechanism for the community trust that the program needs to operate. An agency that briefs its civilian oversight body before launching the AI program, provides regular performance reports that include the integrity metrics, and responds to oversight requests with the same audit trail it would produce for a court, is an agency that is using the oversight structure correctly.

The civilian oversight body that has been aligned on the program before a complaint comes in is an oversight body that can evaluate the complaint factually, against the program's documented standards, rather than against a general suspicion that something is being hidden. That is a very different adjudication environment than the one an agency faces when the oversight body hears about the AI program for the first time through a complaint.

One Story Across Three Audiences

The critical structural requirement of the three-audience alignment process is that the core facts do not change. The depth changes. The vocabulary changes. The emphasis changes. The factual account does not.

Command staff hears: "Axon Draft One drafts the narrative from BWC audio. Officers run the footage-grounded verification pass against the recording and the CAD entry before adopting the draft. The gap-fill rate in the pilot was 14 percent; officers caught 96 percent of those before submission. The disclosure coordinator provides the prosecutor's office with AI-use documentation in the agreed format within 48 hours of report submission."

Council hears: "The AI tool drafts reports from body-camera recordings. Officers are required to check every factual claim against the footage before filing. The prosecutor's office has reviewed and accepted our disclosure format. An independent audit can verify our compliance. The program returned an estimated 2.4 hours per officer per week to street patrol."

Community hears: "Officers now have a tool that helps draft the written part of their reports. A human officer still checks every fact against the body camera before the report is official. When AI helped write a report, that is disclosed to the defense attorney, so defendants know. If you want to know whether AI was used in a report related to a specific incident involving you, here is how to request that information."

These are not three different stories. They are the same story told in the language each audience can use. The council member who receives a constituent complaint can say: "The policy requires the officer to verify every fact against the body camera, and that verification is logged. The defendant's attorney is notified when AI was used. If you want more detail, here is the program's public documentation." That answer is possible only because the council member was aligned on the same factual account, in their vocabulary, before the question came in.

Recovering From an Alignment Failure

The deputy chief's council chamber moment is not a unique story. Agencies across the country have launched AI programs and then encountered a moment of public scrutiny for which they were not prepared: a council question, a newspaper investigation, a community complaint, a civil rights organization's report. In many cases, the program was operationally sound. The problem was alignment, not performance.

Recovering from an alignment failure requires acknowledging it explicitly. An agency that tries to manage its way through a council pause without addressing the underlying alignment gap will encounter the same problem again. The recovery process is the same as the initial alignment process, but it is more expensive in time, political capital, and staff energy, and it begins with credibility that has already been spent.

The honest recovery statement is: "We moved too fast on rollout without building the factual foundation with you. Here is the program's complete record to date: the pilot data, the gap-fill log, the verification compliance rate, the disclosure format the prosecutor's office has reviewed, and the audit trail for every report generated during the pilot. We are asking you to evaluate the program on its actual performance, which we should have put in front of you before the pause vote." That statement provides what the audience needed. It is uncomfortable. It is also the only statement that can actually restart the relationship.

Key Takeaways

• Alignment is not agreement. It is a shared factual understanding across command staff, city council, and community about what the agency's AI program does, what safeguards exist, and how problems are caught. Agreement is too high a bar; alignment is the achievable and necessary standard.
• The three audiences require the same factual account at different depths and in different vocabularies. When the facts diverge across audiences, the program's credibility is damaged in ways that take months to repair.
• Command staff alignment requires program-specific operational knowledge: the verification standard, the gap-fill log from the pilot, the disclosure format, and the audit trail. A commander who cannot answer the deposition question has not been aligned.
• The council presentation leads with the problem (30 to 40 percent of shift on paperwork), covers safeguards as an integral program element (not a defensive afterthought), and presents performance data that includes integrity metrics alongside efficiency metrics.
• The King County, Washington, prosecutor's bar on AI-written reports and the EFF's transparency concerns are not obstacles to dismiss; they are the specific concerns that the aligned presentation addresses directly and factually.
• Community engagement is a sustained process: the community receives factual information, has a meaningful mechanism to raise concerns, and receives responses in plain language. The EFF's opacity concern becomes a program strength when the agency makes its operations genuinely transparent.
• Civilian oversight bodies are alignment partners, not adversaries. An oversight body briefed before the program launches can evaluate complaints factually; an oversight body that hears about the program for the first time through a complaint cannot.
• Recovering from an alignment failure is more expensive than building alignment correctly the first time. The recovery requires acknowledging the failure explicitly, providing the complete factual record that should have been provided earlier, and restarting the alignment process with spent credibility as the starting position.