Sergeant Elena Vasquez had twenty-two officers in her unit and a problem she could not solve with a policy memo: half of them thought the AI drafting tool was going to write their reports for them and the other half thought it was going to get them fired. Neither group was operating it correctly, and both groups had arrived at their position without any formal instruction on what the tool actually did, what its legal implications were, or what a responsible adoption of an AI draft looked like under oath in a deposition. The agency had purchased the Axon Draft One license, rolled it out to the body-worn camera (BWC, the recording device clipped to each officer's uniform) platform, and issued a two-paragraph policy statement. The training consisted of a fifteen-minute product walkthrough video that the vendor had produced. Vasquez had watched it. It explained how to generate a draft. It did not mention Brady v. Maryland, which requires prosecutors to disclose evidence favorable to the defendant, or Giglio v. United States, which extends disclosure obligations to impeachment evidence about officer witnesses. It did not mention that officers spend 30 to 40 percent of their shift on paperwork and that the tool was tested to reduce report-writing time by 82 percent. It did not explain what to do when the draft said something the footage did not show. In short, it taught the mechanics but not the meaning, and that gap was producing officers who were either over-trusting or under-utilizing a tool that, handled correctly, would genuinely help them.

What AI Literacy Actually Means in This Context

AI literacy in the public-safety context is not about teaching officers, dispatchers, and records staff to build AI systems. It is not about understanding transformer architectures or probability distributions. It is about something much more specific and much more operationally urgent: understanding what the tool does, what it can get wrong, what the legal consequences of those errors are, and what the officer's professional obligations are when they use the tool's output in an official capacity.

A patrol officer who is AI literate in this sense can explain the following things without consulting a reference card: what Axon Draft One is doing when it generates a draft from BWC audio (transcribing the audio and producing a narrative from that transcription, filling gaps in the audio with statistically likely descriptions); what a gap-fill is and why it is a problem (the model invents a detail to complete the narrative where the audio did not provide one, and that invented detail may reach a sworn report as if it were a verified fact); what Brady and Giglio mean for the report they are about to submit (if the AI draft contained an error that was corrected, that correction history may be disclosure material the prosecution must provide to the defense); and what "you are the author" means in practice (the officer reviews, corrects, and adopts the draft as their sworn account, and "the computer wrote it" is not an answer in a deposition).

A dispatcher who is AI literate can explain what the AI call-classification tool does with the incoming audio, what the difference is between a suggested call type and an assigned call type, and why the telecommunicator owns the final priority decision regardless of what the AI suggests. A records supervisor who is AI literate can explain why AI-assisted redaction is a first pass rather than a final product, what the difference between over-redaction and under-redaction means under open-records statutes, and what the Criminal Justice Information Services (CJIS, the FBI's security policy governing criminal justice data) obligations are for the AI use log and the draft archive.

None of these explanations require a background in machine learning. They require a specific, role-calibrated literacy that the agency must build deliberately, because the vendor's walkthrough video will not provide it and the policy memo will not maintain it.

An AI-literate officer is not one who can use the tool. It is one who can explain to a prosecutor, a defense attorney, and a judge exactly what the tool did and exactly what they did to verify it. That explanation is the literacy that matters.

Why Trust Must Come First

The operational failure in Vasquez's unit was a trust failure before it was a literacy failure. Half her officers did not trust the tool. The other half trusted it too much. Both groups were operating from a position of uncertainty about what the technology actually was and whether the people deploying it were being straight with them about the risks.

This pattern is not surprising. Law enforcement agencies have a long history of deploying technologies, from digital fingerprint matching to predictive policing tools, with more confidence in the vendor's claims than in the officer's lived experience of the tool's limitations. Officers who have watched a technology deployed with optimistic promises and then encountered its failure modes in the field are justified in their skepticism. They are also, in some cases, justified in their resistance, because the technology's failure mode is their personal professional exposure. When an AI-drafted report contains an error and that error reaches a sworn narrative, the officer is the one in the deposition. The vendor is not.

Trust-first workforce development does not mean persuading skeptical officers to trust AI tools uncritically. It means giving them enough real information about what the tools do, what they get wrong, and what the verification standard is, that their trust is calibrated rather than blind. A skeptical officer who has been given accurate information about the gap-fill failure mode, who knows what a footage-grounded verification pass looks like, and who understands that the disclosure protocol protects them as well as the defendant, is not a tool-resistant officer. They are a tool-responsible officer, and that is the profile the agency needs.

The Trust Deficit and How It Develops

The trust deficit in AI tool adoption typically develops from three sources. The first is information asymmetry: officers learn about the tool from the vendor's promotional materials rather than from honest internal briefings that include the limitations. When the tool's first failure mode surfaces in practice, it feels like a surprise they were not prepared for, which damages trust more than if they had been told about the failure mode in advance.

The second is accountability displacement anxiety: the sense that the agency is using AI to create documentation faster without investing proportionally in the review infrastructure that keeps that faster documentation accurate and defensible. Officers who suspect that AI tools are being deployed to increase throughput without increasing oversight are not wrong to be concerned. That is a real risk in agencies where the governance triangle (the Agency AI Lead, the Disclosure Coordinator, and the Review Cadre) has not been established alongside the tool deployment.

The third is career uncertainty: the worry that AI tools will eventually reduce the need for officers in administrative roles or that demonstrating proficiency with AI tools will be used to set productivity benchmarks that make the job harder rather than better. This concern deserves a direct answer, not a dismissal. The honest answer is that AI tools designed for report drafting return time to the officer, specifically the 30 to 40 percent of a shift that currently goes to paperwork, and that time goes back to patrol, investigations, and community engagement. The job is not smaller. It shifts toward judgment and away from transcription. Whether individual agencies actually use the time that way is a leadership and resource decision, and officers are right to pay attention to whether their agency is making that commitment.

Building Trust Through Transparency About Limitations

The most effective trust-building intervention is honest communication about limitations before the limitations are encountered in the field. When the agency's AI literacy program leads with the gap-fill failure mode, when it teaches the three ways an AI-drafted report can contain an error (the gap-fill detail, the softened fact, and the invented quote) and the verification steps that catch each one, it is telling officers: we know this tool has limitations, we are giving you the tools to catch them, and we are not asking you to take on a risk we have not told you about. That transparency builds more trust than a vendor walkthrough that presents the tool as a solution.

The Electronic Frontier Foundation (EFF, a civil-liberties organization that monitors AI use in law enforcement) has raised concerns about agencies deploying AI report-writing tools without transparent disclosure to the public about how those tools are used, how errors are caught, and how the outputs are documented in discovery. The EFF's concerns are about public trust, but they mirror the internal trust issue exactly: when people, whether they are community members or sworn officers, do not have accurate information about what a tool does and what its limitations are, they cannot make calibrated decisions about how to relate to it. The transparency that builds community trust and the transparency that builds officer trust are the same transparency. Agencies that lead with honest, complete information about the tool do better on both dimensions.

A Literacy Curriculum: From Patrol to Command

AI literacy in a public-safety agency is not one curriculum. It is several, calibrated to the specific role and the specific decisions that role holder will be making in the AI-assisted workflow. The patrol officer, the dispatcher, the records supervisor, the detective, the training officer, and the command-staff executive all need to understand AI tools in the context of their specific responsibilities. A curriculum that teaches the same content to everyone delivers too little to the people who need depth and too much abstract policy to the people who need operational specificity.

Patrol Officer Curriculum

The patrol officer curriculum has four core components, and all four must be present for the officer to operate the tool responsibly.

The first component is the tool's function: what Axon Draft One (and comparable tools from other vendors) does, specifically. The BWC records the audio of the incident. The platform transcribes that audio. The AI model generates a narrative draft from the transcription. The draft is a starting point, not a finished product. The officer is the author of the finished product, not the tool.

The second component is the failure modes: what the tool gets wrong and why. The gap-fill detail (the model invents a specific fact to fill an audio gap), the softened fact (the model describes an action in less specific or less forceful terms than the footage supports), and the invented quote (the model produces a direct quotation that does not appear in the audio). Each of these is a Brady problem if it reaches a sworn report uncorrected. Each is catchable with a footage-grounded verification pass.

The third component is the verification standard: how to run the footage-grounded verification pass. Open the draft and the footage side by side. Go to each factual claim in the draft and verify it against the footage at the relevant timestamp. If the footage supports the claim, mark it verified. If the footage contradicts the claim, correct the draft to reflect the footage and document the correction. If the footage does not address the claim, flag the claim as based on personal observation (if it is) or remove it (if it is not). The verification pass is not a read-through. It is an active, timestamped comparison.

The fourth component is the authorship and disclosure obligation: what adopting the draft means legally, and what the disclosure notation must say. When the officer certifies and submits the report, they are adopting it as their sworn account. "The AI wrote it" is not a defense, a disclaimer, or a qualification. It is the answer that ends a career. The officer is the author. The disclosure notation documents what the tool was used for, that the officer reviewed the output, and that the officer is the author of the adopted report. That notation is the mechanism that meets the King County, Washington prosecutor's concern about AI-written reports head-on: it provides exactly the accountability and transparency that the King County bar was designed to ensure existed before a report reached prosecution.

Dispatcher Curriculum

The dispatcher curriculum is anchored in a single principle: the telecommunicator owns the priority. The AI may suggest a call type. The AI may surface prior calls at the address. The AI may generate the initial computer-aided dispatch (CAD, the system that tracks calls, units, and dispatch decisions) entry text. None of those functions transfers the priority decision to the AI. The telecommunicator hears the caller's voice. They hear the fear, the ambient sound, the information the caller is giving that does not fit the words. That hearing is a human input the AI does not have, and it is often the deciding input in a correctly classified call.

The dispatcher curriculum covers what the AI is doing with incoming audio, how the suggested call type is generated and what it is based on, how to override the suggested type and document the override, and what the consequence is of accepting an AI-suggested call type for a call the dispatcher's judgment would have classified differently. A call that should be a domestic violence emergency classified as a noise disturbance is a delayed response to a life-safety incident. The word "misclassification" understates the consequence. The dispatcher curriculum should use real or near-realistic scenarios to make that consequence concrete, not hypothetical.

Records Staff Curriculum

The records staff curriculum covers the specific AI functions relevant to the records and redaction workflow: AI-assisted redaction, public-records request processing, and the maintenance of the AI use log. Records staff need to understand why AI-assisted redaction is a first pass rather than a final product (the consequences of a missed face or plate under open-records statutes), how to conduct the human review of AI-flagged redactions, and how to handle the situations where the AI over-redacted (blocking information that should have been released) or under-redacted (releasing information that should have been protected).

Records staff also need to understand CJIS obligations as they apply to the materials they are managing: the AI draft archive, the verification log, the AI use log. These are criminal justice records. They are not administrative documents that can be retained and disposed of on the agency's general records schedule. They have their own retention and access requirements, and the records staff who maintain the RMS are the people who will be responsible for producing those records in response to a discovery request or an audit.

Command Staff Curriculum

The command-staff curriculum is different in character from the operational curricula. Command staff do not need to know how to run a footage-grounded verification pass. They need to know what it is, why it matters, and whether the agency's workflows are designed to ensure it is being done. They need to be able to answer the city council's question about what the agency's AI program looks like, what the verification standard is, and what happens when the AI gets something wrong.

Command staff need to understand the contractual exposure that comes with bundled, multi-year, sole-vendor AI contracts. Deals in the range of $45 million over ten years, encompassing cameras, cloud storage, and AI tools in a single package, create dependencies that outlast the policies governing them. A ten-year contract signed in 2025 will span multiple mayoral administrations, multiple police chiefs, multiple iterations of the AI tool, and potentially multiple changes in the legal framework governing AI use in law enforcement. Command staff who understand that risk are equipped to negotiate contract terms that protect the agency's ability to adapt: advance notice of model updates, audit rights over vendor data handling, exit provisions that do not require forfeiting the hardware investment.

Command staff also need to understand the trust-and-accountability narrative: how to tell the agency's AI story to city council, to community groups, to oversight boards, and to the media in a way that is accurate about both the opportunity and the risk. The time the AI tools return to officers, the 30 to 40 percent of a shift currently spent on paperwork, and the verification standard that preserves the evidentiary value of the reports those tools help produce, are two parts of the same story. Command staff who can tell both parts together, including the EFF's concerns and the King County precedent as context rather than obstacles, are the ones who build and maintain the community trust that is the program's long-term operating environment.

Delivering the Curriculum: Formats That Work

Curriculum content without a delivery mechanism does not produce literacy. The delivery formats for AI literacy in a public-safety agency need to work within the constraints of shift work, high turnover, and the operational pace that makes long classroom sessions impractical. Several formats have proven effective in analogous training contexts and translate well to AI literacy.

Scenario-Based Roll Call Training

Roll call training is the most consistent access point for patrol officers and dispatchers in most agencies. The format works for AI literacy when the content is scenario-based: a short, specific case where an AI error had a consequence, and a five-to-seven-minute discussion of what the verification step would have caught and how to run it. The scenario should be drawn from real cases where possible (anonymized), or from the near-realistic hypotheticals that the Review Cadre develops from the error data they are collecting in quality review. The goal is to make the failure mode concrete and the fix specific, not to cover policy in the abstract.

Roll call training should recur quarterly, with different scenarios each time, to keep the verification standard in the operational memory of officers who are using the tool daily. A one-time rollout training that happened when the tool was deployed in 2024 is not AI literacy in 2026. Literacy requires maintenance.

Structured Field Training Integration

Field training officers (FTOs) are the most influential trainers in a patrol officer's formative period. If FTOs are not trained in the AI literacy curriculum and not incorporating verification-pass practice into their training with new officers, new officers will learn their AI workflow habits from whoever they ride with, which may or may not include the habits the agency has defined as its standard. AI literacy must be integrated into the field training curriculum explicitly, with specific competencies that new officers must demonstrate before the training period ends: running a footage-grounded verification pass on a sample report, correctly flagging an AI-assisted report in the RMS, and being able to articulate the Brady disclosure obligation in plain language.

Tabletop Exercises for Supervisors and Command

Supervisors and command staff benefit from tabletop exercises: structured scenarios where a simulated AI-related incident (a suppression motion citing a gap-fill, a defense discovery request for the original AI draft, a community complaint about AI use in a use-of-force report) is presented and the group works through the agency's response. The tabletop format surfaces gaps in the workflow and disclosure protocols in a low-stakes environment rather than in an actual case. It also builds the command staff's confidence in explaining the program, because they have rehearsed the conversation.

A tabletop exercise calibrated to this lesson's content might present the following scenario: a detective adopted an AI-assisted case summary that contained an invented quote attributed to a witness. The error was not caught in the verification pass. The case is now in trial, and defense counsel has requested the original AI output in discovery. The exercise asks: what is the agency's disclosure obligation? Who handles the discovery request? What is the Review Cadre's finding going to show? How does the AI Lead brief the prosecutor's office? What does the public statement look like if the story reaches the press? Working through those questions in advance is the command-staff version of the footage-grounded verification pass: it is catching the gap before the incident reaches the courtroom.

Annual Certification and the Audit Trail

Every officer and staff member who uses AI tools in their daily work should complete an annual certification that documents their current AI literacy. The certification is not primarily for the individual. It is for the agency's audit trail. When a defense attorney asks "did your officers receive training on the AI tool's limitations and the verification standard?", the answer should be "yes, here is the signed training record for every officer who used the tool in the relevant period." That record is the agency's evidence that the literacy program was real, not aspirational.

Annual certification should include a brief assessment: not a high-stakes exam, but a structured set of questions that confirm the officer can identify the three failure modes, describe the footage-grounded verification pass, and articulate their disclosure obligation. The assessment results feed the AI Lead's training effectiveness data, which is part of the program's performance dashboard. If officers are consistently missing questions about a specific failure mode, that is a training gap, not a personnel gap, and the curriculum needs to address it.

The Trust-First Culture: From the Inside Out

AI literacy training delivers information. Culture delivers behavior. The difference between an agency where officers consistently run rigorous verification passes and one where they do not is not primarily a difference in the information those officers received. It is a difference in whether the culture tells them their verification work matters, that supervisors are checking it, that the agency is investing in it as seriously as it invested in the tool itself.

A trust-first culture is built by a few specific leadership behaviors that reinforce it consistently. When a Review Cadre member catches a systematic error in the AI tool's output for a specific call type and the finding is acknowledged at a command briefing, that acknowledgment tells the agency: quality review is valued, not just required. When a supervisor spots a well-documented verification correction in an officer's report and notes it in a performance conversation as the right practice, that recognition tells the officer: the work of the verification pass is seen and valued. When the AI Lead presents the disclosure log and the error data to city council as evidence of the program's accountability posture, that presentation tells the community: the agency is taking the oversight seriously, not just the efficiency.

The trust-first culture is also built by honest acknowledgment of the tool's limitations in every public forum where the agency discusses AI. The EFF's concerns about transparency are not a threat to dismiss. They are a prompt to be specific: here is what the tool does, here is what it gets wrong, here is how we catch it, here is the disclosure protocol. That specificity is the answer to the transparency concern, and it is also the behavior that builds the internal culture where officers feel safe raising concerns about AI outputs they are uncertain about, rather than submitting uncertain reports because the system rewards speed over accuracy.

Sergeant Vasquez's twenty-two officers needed the same thing the community needed, and the same thing the prosecutor's office needed: accurate information about what the tool does, honest acknowledgment of what it gets wrong, and a clear organizational signal that the verification work is valued, not just tolerated. That is what a trust-first AI literacy program delivers. It is also the foundation on which a public-safety AI program builds the durability to survive the first case challenge, the first oversight audit, and the first community controversy, and emerge from each one with its credibility intact.

Key Takeaways

• AI literacy in public safety is role-calibrated, not uniform. Patrol officers, dispatchers, records staff, and command staff each need a curriculum anchored in their specific decisions and their specific accountability exposure. A vendor walkthrough video is not a literacy program.
• The gap-fill failure mode, the softened fact, and the invented quote are the three specific ways an AI-drafted report can contain an error. Officers who cannot name these failure modes cannot catch them, and uncaught failure modes become Brady problems.
• Trust-first means building calibrated trust, not blind trust. Officers who receive accurate, complete information about the tool's limitations, the verification standard, and the disclosure obligation before encountering the tool's errors in the field are the ones who operate it responsibly.
• The accountability displacement anxiety is real and deserves a direct answer: AI tools return time to officers, specifically the 30 to 40 percent of a shift currently spent on paperwork. Whether that time goes back to patrol, investigations, and community engagement is a leadership decision the agency must make explicitly and keep.
• Roll call training, field training integration, command-staff tabletops, and annual certification are the four delivery formats that make AI literacy operational rather than aspirational. Each format has a specific function, and all four are necessary for the literacy to reach every role in the agency.
• Annual certification provides the audit trail the agency needs when a defense attorney asks whether officers were trained on the tool's limitations. Without that documentation, the answer is an assertion. With it, the answer is a record.
• The trust-first culture is built by leadership behaviors: acknowledging Review Cadre findings in command briefings, recognizing good verification practice in performance conversations, and presenting the disclosure log to city council as evidence of accountability. Information builds literacy. Culture builds behavior.
• The EFF's transparency concerns and the King County prosecutor's decision to bar AI-written reports are not obstacles to the program. They are the design requirements. An agency whose literacy program produces officers who can explain the tool, the verification standard, and the disclosure protocol to any audience, including a skeptical one, is operating the program the way it should be operated.