Chief David Kim stood at the podium in the city council chamber with two very different audiences in the room. On the raised dais in front of him sat seven council members who had read the budget projection: a proposed AI reporting suite that, if the efficiency numbers held, would return the equivalent of 35 full-time officer-equivalents of patrol capacity without a single new hire. Behind him, filling most of the public gallery, sat representatives from three community advocacy organizations, two of whom had already submitted written public comment opposing the initiative on civil-liberties grounds. Both audiences wanted to hear from the chief. They wanted different things. Chief Kim's task was to tell one honest story that served them both, because the story that worked only for the council would not survive the community, and the story that worked only for the community would not survive the council's budget committee.

Why One Story Serves Both Audiences

The temptation, in a room divided like the one Chief Kim faced, is to tell each audience what they want to hear. Tell the council about the efficiency numbers. Tell the community about the safeguards. Keep them separate. The problem is that this strategy always collapses, usually at the moment a community member files a public-records request for the council presentation slides and finds that the slides presented to the council contained no mention of the concerns the chief addressed at the community meeting, or vice versa.

The right approach is different and more durable: one narrative that is honest about both the opportunity and the accountability structure. This approach does not require splitting the difference between two positions. It requires understanding that both positions are responding to real facts. The efficiency gain is real. Officers spend 30 to 40 percent of every shift writing reports. A tool that returns even half of that time to patrol has a genuine, measurable operational impact. The civil-liberties and accountability concerns are also real. The King County, Washington prosecutor's office barred AI-written police reports when the controls were not in place. The Electronic Frontier Foundation (EFF) has raised substantive transparency concerns about AI police report tools that represent a legitimate constituency. Brady v. Maryland (the 1963 Supreme Court decision requiring disclosure of exculpatory evidence) and Giglio v. United States (the 1972 decision requiring disclosure of impeachment evidence) frame AI use in sworn report production as a constitutional matter.

An honest narrative that presents both truths is not a compromise. It is the stronger position. A chief who can explain why the King County bar happened, what the agency's program does differently, and how that difference protects the community and the officer at the same time, is a chief who has done the governance work. The advocacy organizations behind Chief Kim could not dismiss what they could not characterize as spin.

The efficiency story and the accountability story are not competing narratives. They are the same story told completely. Tell the incomplete version and you lose the audience you did not tell it to.

Building the Council Case: Efficiency, Accountability, and Fiscal Return

City council members are accountable to the same taxpayers who want police visibility on their streets and value on their tax dollar. The council case for AI in public safety is most effective when it is built on three interconnected elements: the documented efficiency return, the specific accountability structure that protects it, and the total cost of ownership that produces an honest return-on-investment calculation.

The Efficiency Return

The efficiency case begins with a fact most council members do not fully appreciate: officers spend 30 to 40 percent of every shift writing reports. For a 180-officer patrol force, that is the equivalent of 54 to 72 officers whose available patrol hours per shift are consumed by documentation. AI-assisted report drafting tested with tools like Axon Draft One has shown up to an 82-percent reduction in report-writing time under controlled conditions. Operational results in deployed agencies run lower, from approximately 45 to 65 percent, once the verification pass that proper accountability requires is included. At 55 percent reduction in operational conditions, and accounting for the verification pass adding approximately 15 minutes per report, the net gain in a 180-officer force is the equivalent of approximately 25 to 30 officer-equivalents redirected from documentation to patrol, investigations, and community engagement.

That is not a small number. It is the equivalent of a patrol staffing increase that no city budget could fund through hiring alone. And it is achievable without the multi-year ramp of a new hire: no academy seat, no field training program, no probationary period. The efficiency gain begins in the first operational quarter of phase-two deployment and compounds as more officers complete the training program.

The council case should also present the fiscal case for the records and phase-one components. AI-assisted redaction of body-worn camera (BWC) footage for public-records responses currently takes a records unit between one and two weeks for a complex request involving hours of footage. AI assistance reduces that to one to two days. For an agency receiving fifty to one hundred complex records requests per month, that represents a significant backlog reduction, and backlog compliance with open-government statutes carries real legal risk. Court-ordered record releases that fail to meet deadline produce liability. AI assistance in records reduces that liability without adding records staff.

The Accountability Structure

This is where chiefs most often underperform in council presentations: they present the efficiency numbers and then add the accountability section as a footnote, something to cover them legally rather than something to present as equal in importance to the efficiency numbers. The accountability structure is not a footnote. It is the reason the efficiency gain is sustainable.

The accountability structure has four elements that belong in every council presentation at equal weight.

First, the authorship standard. Every report drafted with AI assistance is adopted by a named officer who has verified every factual claim against the body-worn camera footage, the computer-aided dispatch (CAD) entry, and the case record before signing it as their sworn account. "The computer wrote it" is never an acceptable answer in a deposition. The officer is the author. This is not a policy preference. It is the operational fact that makes the report admissible evidence.

Second, the verification standard. The verification pass is not a read-through. It is active checking: the officer opens the footage, goes to the relevant timestamp, and confirms what the AI draft says happened is what the camera actually recorded. Any claim the footage does not support is corrected before submission. This catches the gap-fill, which is the specific failure mode where the AI invents a plausible detail to complete a narrative when the audio has a gap.

Third, the disclosure standard. The agency's disclosure policy, developed in coordination with the local prosecutor's office, ensures that AI involvement in producing a report is noted in discovery packages provided to the defense. This meets the Brady obligation head-on rather than hoping the defense never asks about the AI's role. The Criminal Justice Information Services (CJIS) Security Policy governs how criminal justice data is handled throughout this process, and those obligations stay with the agency, not the vendor.

Fourth, the monitoring and audit standard. The AI program is actively monitored. A random sample of AI-assisted reports is reviewed against the footage each week to verify that the verification standard is being met, that the error and correction rates are within acceptable ranges, and that systemic platform errors are identified and escalated to the vendor. The monitoring data is the evidence base for future phase decisions and for any oversight review.

Total Cost of Ownership

The fiscal presentation to council should be complete. The vendor's headline contract price is not the program cost. The program cost includes policy development, training, infrastructure upgrades required for CJIS compliance, legal review, and the ongoing monitoring program. Bundled, multi-year, sole-vendor contracts covering cameras, cloud storage, and AI in a single agreement can run approximately $45 million over up to ten years. Those contracts carry lock-in risk: an agency that signs a ten-year sole-vendor agreement at the beginning of a rapidly evolving technology period may find itself committed to a platform that has been superseded, or to a vendor whose practices have changed in ways the agency cannot accommodate. Exit rights and data portability should be in the contract before signing. Council should understand both the headline price and the full commitment.

The return side of the calculation should be equally specific. Officer time returned to patrol has a dollar value: the average cost of a fully-loaded officer-year (salary, benefits, equipment) divided by the percentage of shift time recovered. In most jurisdictions this calculation produces a return figure that makes a significant AI investment financially compelling even in a single year. Present the complete calculation, with the assumptions stated, so the council can evaluate it. A council that understands the assumptions is a council that can defend the investment to their constituents.

Building the Community Case: Transparency, Safeguards, and Accountability

The community case requires a different structure from the council case, but not a different substance. The community does not need to be persuaded that time savings matter. They need to be persuaded that the safeguards are real, not performance. The organizations in Chief Kim's public gallery had seen technology programs described with accountability language that turned out to be aspirational. They were right to be skeptical. The chief's task was to answer the skepticism with specifics, not with reassurance.

The Transparency Commitment

The first and most important commitment is disclosure: when AI is used to assist a report, that fact is disclosed in the discovery package provided to the defense. This is not a concession to critics. It is what Brady and Giglio require. Telling the community that AI involvement is disclosed to the defense in every case is telling them that the courts, not just the agency, will know when AI was involved. That is a more credible accountability mechanism than any internal policy, because it is subject to adversarial scrutiny.

The second commitment is transparency about what AI does and does not do. In a report context, AI drafts a narrative from the body-camera audio. It does not decide guilt. It does not determine what charges are filed. It does not replace the officer's sworn account. The officer reads the draft, checks every fact against the footage, corrects what the AI got wrong, and adopts the result as their sworn account. The AI is a drafting tool. The officer is the author. This framing matters because the advocacy organizations' concern is often based on a misunderstanding of what AI involvement means: they imagine the AI making decisions. Correcting that misunderstanding accurately, without minimizing the real concerns, is a service to the conversation.

Addressing the Legitimate Concerns Directly

The EFF's concerns about AI police report tools are substantive and documented. The specific concerns center on transparency: community members do not know when AI is involved in producing the evidence used to prosecute them, and they cannot evaluate the accuracy or bias of AI tools they cannot see. These are legitimate concerns. The answer to them is not dismissal; it is specific documentation.

Chief Kim's presentation to the community addressed each concern with a specific mechanism. On transparency: AI involvement is disclosed in every discovery package, and the agency's AI use policy is a public document available for review. On accuracy: the verification standard requires every factual claim to be confirmed against the footage before the report is adopted, and the monitoring data from the pilot program is available to community oversight bodies. On bias: the agency committed to a periodic audit of AI-assisted reports against similar incidents without AI assistance to detect any systematic language patterns that should not be present.

The King County prosecutor's bar on AI-written reports was addressed directly, not avoided. "The reason King County barred those reports is that the controls we are putting in place were not there. Our verification standard, our disclosure policy, and our prosecutor coordination are specifically designed to meet the test that King County's bar defined." That is a more credible answer than pretending the bar did not happen or dismissing it as irrelevant.

On the question of what happens if something goes wrong: the agency has a written error escalation pathway. An officer who discovers a material AI error in a draft documents it, reports it to a supervisor, and it enters the error-tracking system that feeds back to the vendor. Cases in which an AI error was corrected before submission are documented and available for oversight review. Cases in which an error reaches a final report are treated as serious incidents that trigger a review of the training and monitoring program.

The Community Advisory Mechanism

The strongest community accountability commitment a chief can make is an ongoing one: a community advisory structure, not a one-time briefing. Chief Kim's proposal included a civilian advisory board, consisting of representatives from each of the advocacy organizations present, with quarterly access to the AI program's monitoring data, error rates, correction rates, and any changes to the disclosure policy. This was not a gesture. It was a structural commitment that the agency would continue to be accountable to the community on AI, not just at the announcement but at every phase transition and every significant policy change.

This kind of structural commitment is the difference between a briefing and a relationship. A briefing can be received and opposed. A relationship, structured around access to real monitoring data, requires the advocacy organization to engage with the evidence rather than with the narrative. If the monitoring data shows the program is working as the agency said it would, the relationship produces credibility. If the monitoring data reveals a problem, the advisory structure gives the community the standing to raise it before it becomes a crisis.

The One Honest Narrative: Writing and Delivering It

Chief Kim's presentation ran forty-five minutes. The first fifteen minutes covered the operational case: the documentation burden, the specific efficiency return, the fiscal analysis, and the roadmap. The next fifteen minutes covered the accountability structure: authorship, verification, disclosure, and monitoring. The final fifteen minutes were dedicated to the community engagement plan, the advisory structure, and direct answers to each of the written public comments the advocacy organizations had submitted.

He did not lead with the 82-percent benchmark. He led with the 30-to-40-percent documentation burden and the operational reality it creates for officers who should be available to the community they serve. He did not minimize the EFF's concerns or the King County bar. He explained them, attributed them correctly, and then explained specifically how the agency's program addressed the governance failures that produced them.

The council approved the phase-one budget unanimously and asked to receive monitoring data before approving phase two. Two of the three advocacy organizations submitted written letters after the meeting noting that they remained cautious but would participate in the advisory structure. The third organization continued its formal opposition but narrowed its objections from "oppose AI in policing" to specific questions about the bias audit methodology, which the agency addressed in the first quarterly advisory meeting.

That outcome is what the one honest narrative produces. Not universal endorsement: some opposition is legitimate and will remain. But the kind of engaged, evidence-based accountability relationship that makes the program governable rather than embattled. The council understands the investment. The community understands the safeguards. The prosecutor's office understands the disclosure standard. The opposition has been heard and has specific questions to engage with rather than a narrative to oppose.

Talking Points for the Hard Questions

Command staff preparing a council or community presentation should have prepared answers to the five questions that will come from every audience that has thought carefully about AI in policing.

"What happens if the AI makes a mistake that sends an innocent person to jail?" The verification standard is the answer. Every factual claim must be confirmed against the footage before the report is adopted. The officer, not the AI, is the author of the sworn account. A mistake in the AI draft that is caught before adoption is a caught and corrected draft, not a criminal outcome. A mistake that is not caught before adoption is an officer accountability failure, and the agency's training and monitoring program exists to prevent that.

"Are you just trying to save money at the expense of accuracy?" No. The efficiency gain is real and important: officers spending 30 to 40 percent of a shift on paperwork are not available for the community work that technology budgets are supposed to enable. The accountability standard requires that efficiency to be earned by maintaining report accuracy. The monitoring program tracks both the time savings and the correction rate. If the correction rate shows an accuracy problem, the deployment stops.

"Who sees the AI output before the officer uses it?" Nobody else needs to. The AI draft is generated from the officer's body-camera audio, reviewed by the officer against the footage, corrected, and adopted. It does not go to a supervisor for AI-specific approval. The officer's authority and responsibility over their own sworn report are not diminished by the fact that a draft was generated. The verification standard is more rigorous than reading a draft and thinking it sounds right. It is footage-grounded active checking.

"What does CJIS mean for this program?" CJIS is the Criminal Justice Information Services Security Policy, the FBI's framework governing how criminal justice data must be handled. Every data flow between the body-camera platform, the AI system, and the records management system has been mapped and verified for CJIS compliance. Those obligations stay with the agency. The vendor's CJIS certification is a necessary but not sufficient condition: the agency independently verified compliance before deployment.

"If King County banned AI reports, why should we allow them?" Because the reason King County banned them was the absence of the controls we are implementing. Their bar was a governance verdict, not a technology verdict. The verification standard, the disclosure policy, the prosecutor coordination, and the monitoring program are specifically designed to meet the test that King County's bar defined. We welcome the prosecutor's scrutiny, because a program that can withstand that scrutiny is a program that serves the community correctly.

Key Takeaways

• The efficiency story and the accountability story are one honest narrative, not two competing ones. A chief who tells them together is in a stronger position than one who tells them separately to different audiences.
• The council case rests on three elements: the documented efficiency return (30 to 40 percent of a shift on paperwork, 45 to 65 percent operational time reduction), the accountability structure (authorship, verification, disclosure, and monitoring), and the honest total cost of ownership that includes remediation and contract terms alongside the headline price.
• The community case requires specific mechanisms, not reassurances. Disclosure in every discovery package, public AI use policy, access to monitoring data, and an ongoing advisory structure are more credible than general commitments.
• The King County prosecutor's bar on AI-written reports must be addressed directly, not avoided. The answer that survives scrutiny is: "The controls we have put in place specifically address the governance failures that produced that bar."
• Brady v. Maryland and Giglio v. United States frame AI disclosure as a constitutional matter. Telling the community that AI involvement is disclosed to the defense in every case is telling them that adversarial scrutiny, not just agency policy, governs the tool.
• The EFF's transparency concerns are substantive and should be answered with specific documentation: the public AI use policy, the monitoring data available to oversight bodies, and the bias audit commitment.
• Bundled, multi-year, sole-vendor contracts can reach approximately $45 million over up to ten years. Council should understand the lock-in risk, the exit rights, and the data portability provisions alongside the efficiency projection.
• An ongoing community advisory structure with access to real monitoring data is the difference between a one-time briefing and an accountable relationship. Opposition that has specific questions to engage with is more governable than opposition that has only a narrative to oppose.