The chief of a mid-sized department sat across from his technology vendor at a budget presentation in the spring of 2025. The slide deck was polished. It showed a drone-swarm concept, a predictive routing engine, a real-time behavioral-analysis feed piped into the computer-aided dispatch (CAD, the system that tracks active calls and unit assignments) console, and an automated interview-flagging tool that would surface "anomalous" statements in recorded interrogations. The vendor called it "the future of public safety operations." The chief called his legal counsel that afternoon. By morning he had a list: three capabilities that were genuinely defensible, one that was promising but needed a pilot, and two that he would not touch without a city-council ordinance and a civil-liberties review. He did not make that list because he was anti-technology. He made it because he understood, at the highest-stakes level an executive can understand it, that pursuing the wrong use case is not neutral. It creates exposure, erodes community trust, invites legislative action, and in the worst cases produces outcomes that harm the people the agency is supposed to protect.

Why Use-Case Identification Is an Executive Discipline

At the officer and practitioner levels covered in earlier parts of this program, the relevant question is "am I using this tool correctly?" At the executive level, the relevant question is "should we use this tool at all, and under what conditions?" That is a different intellectual task. It requires a structured way of thinking about where AI genuinely helps, where the civil-liberties cost is low enough that the help is worth pursuing, and where the potential value is real but the governance requirements are so significant that the agency must build the governance before it deploys the tool, not after.

This discipline matters more in 2026 than it did five years ago, because the pace of vendor innovation has outrun the pace of most agencies' policy-making. A vendor will offer a capability in a sales conversation that the agency's policy does not address, the procurement office does not know how to evaluate, the prosecutor has not blessed, and the city council has never discussed. The executive who says yes in that sales conversation has made a governance decision by accident. The executive who says "not yet, and here is what would have to be true before I say yes" is doing the job right.

There is also a structural reason to take use-case identification seriously at the executive level: the decisions compound. An agency that deploys AI report-writing correctly, with verification, disclosure, and an audit trail, builds credibility with prosecutors, with oversight bodies, and with the community. That credibility is a political asset. It makes it easier to bring the next proposal to the city council with a track record, not just a promise. An agency that deploys the wrong use case first, or deploys the right use case without adequate governance, spends that credibility in a way that makes every subsequent proposal harder. The order matters.

The Three Categories of Use Case

For practical purposes, executive use-case assessment produces three categories. Understanding them is the foundation of the identification discipline.

The first category is high-value, lower-scrutiny. These are applications where the efficiency gain is well-documented, the civil-liberties concerns are manageable with existing practice and policy, and the failure modes are recoverable. AI-assisted report drafting from body-worn camera (BWC, the recording device clipped to an officer's uniform) footage, with a documented verification pass and disclosure policy, falls here. The 82% reduction in report-writing time documented in testing of Axon's Draft One tool is a concrete, audited figure. Officers currently spend 30 to 40% of every shift on paperwork. Returning a meaningful share of that time to patrol and investigations is a legitimate public safety benefit. The failure modes, which are the gap-fill error and the unverified narrative, are serious, but they are manageable with training, verification standards, and the governance structure this program describes. A prosecutor-approved disclosure protocol addresses the King County-style objection before it arises. This category is where responsible deployment starts.

The second category is high-value, high-scrutiny. These are applications where the potential public safety value is real and well-documented, but where the civil-liberties concerns are significant, the failure modes may not be recoverable (a wrongful arrest, a chilling effect on protected speech or association, a discriminatory pattern baked into an automated system), and the governance requirements are substantial. Drone-as-first-responder (DFR) programs and real-time crime centers (RTCCs, which aggregate multiple live camera feeds, license plate reader data, and other sensor inputs for monitored response) are the clearest examples in 2026. The value is not in dispute. Drones that reach a scene before a patrol unit can improve officer safety and incident documentation in ways that are measurable. RTCCs that correlate camera feeds, license plate readers, and CAD data have reduced response times and improved case closure rates in agencies that have deployed them carefully. The scrutiny is also not in dispute. The Electronic Frontier Foundation (EFF, the digital civil-liberties organization that monitors surveillance technology) has raised substantive concerns about persistent aerial surveillance, mass license plate data retention, and the chilling effects of pervasive monitoring on communities. These concerns are legitimate and must be treated as central to any deployment decision, not as public relations challenges to manage.

The third category is not yet. These are applications where the claimed value is insufficiently documented, the civil-liberties cost is high, the failure modes are severe, or the governance requirements exceed what the agency can credibly provide. Predictive policing tools that assign risk scores to people rather than places, automated behavioral-analysis tools applied to interrogations without independent validation, and facial recognition used for investigative purposes without a documented accuracy floor, probable cause standard, and human-review requirement all belong here for most agencies in 2026. "Not yet" does not mean "never." It means the evidentiary bar for the value claim has not been met, the governance infrastructure has not been built, and the civil-liberties cost-benefit analysis has not been done in a way that the agency can defend publicly. Some of these tools may eventually belong in category two, with the right governance. None of them belong in category one. The executive's job is to know which category a given proposal is in before the contract is signed.

The Responsible-Innovation Line: What It Is and Where It Runs

The phrase "responsible innovation" is used frequently in technology policy contexts, and its meaning has become diluted through overuse. For purposes of this lesson, it has a specific and operational meaning: responsible innovation is the deployment of a new capability at a scale and pace that the agency's governance infrastructure can actually support, with a civil-liberties cost that the agency can honestly defend to the community it serves.

That definition has two components that must both be satisfied. The first is a governance-infrastructure test: does the agency have the policy, the training, the audit capability, and the human-review structure to operate this tool as it is supposed to be operated? Not as the vendor says it can be operated. As it will actually be operated at 2 AM on a understaffed shift when the supervisor is handling something else. The second is a civil-liberties cost-benefit test: can the agency make a public, honest case that the benefit to public safety outweighs the privacy cost, the surveillance risk, and the potential for discriminatory impact? That case must be made to the community, not just to the command staff. If the agency cannot make it publicly, the civil-liberties cost is too high or the benefit case is too weak.

The responsible-innovation line is not drawn by the technology. It is drawn by the governance infrastructure the agency has built and the civil-liberties case the agency can publicly defend.

The Vendor Conversation at Executive Level

Understanding what vendors offer, and how to receive those offers as an executive, is a distinct skill. Vendors in the public safety AI space are sophisticated, well-resourced, and operating in a market where multi-year, multi-million-dollar contracts are the norm. Bundled contracts that pair body cameras, drones, evidence management cloud platforms, and AI reporting tools have reached approximately 45 million dollars and up to ten years in duration in some procurement cycles. The scale of those contracts is not incidental. It creates incentives for the vendor to present an expansive vision of what the technology can do, and for agency leadership to evaluate the technology in the context of a relationship that already exists rather than purely on the merits of the specific proposal.

The executive posture in a vendor conversation about a new capability should be evidence-first. Every performance claim deserves a sourced citation: what study, what agency, what conditions, and what validation method produced this figure? A vendor who says their tool reduces response time by a stated percentage should be asked to produce the study, identify the agency, describe the deployment conditions, and explain what the control condition was. The figure in the slide deck is a marketing output. The underlying study, if it exists, is the relevant document. If the underlying study does not exist, or if it was conducted by the vendor itself without independent validation, that is a signal about the strength of the value case.

The Civil Justice Information Services (CJIS, the FBI division that sets security standards for handling criminal justice data) Security Policy obligations stay with the agency regardless of what the vendor contract says. An agency cannot transfer its CJIS compliance obligation to a vendor by signing a data processing agreement. The vendor can certify that its platform is CJIS-compliant. The agency remains responsible for ensuring that its use of the platform is compliant. That distinction is consequential when a vendor offers a new capability that involves new data types, new processing locations, or new sharing arrangements. The executive who approves a new AI capability without confirming its CJIS compliance posture has created an exposure that may not surface until an audit or an incident.

The Use-Case Assessment Framework for Executives

The following framework gives agency executives a structured way to assess any proposed AI use case before it reaches the procurement stage. It is designed to surface the questions that matter at the governance level, not at the technical level.

Value Clarity

What is the claimed benefit, and is the evidence for it credible? This requires distinguishing between a performance claim from a controlled study, a benchmark figure from a vendor's testing environment, a case study from a single agency with specific conditions, and a theoretical benefit from a design document. Each has a different evidentiary weight. The 82% figure for Axon Draft One's effect on report-writing time comes from testing with officers and has been cited across multiple contexts. It is a benchmark figure that deserves verification in the agency's own environment, not a guarantee, but it has a credible basis. A vendor's slide claiming that their tool "reduces crime by 20%" using their own proprietary study with no peer review has a much weaker evidentiary basis.

The value question also has a deployment-reality component. What will this tool actually do in this agency, with this staffing pattern, this call volume, and this technology infrastructure? A tool that produces strong results at a large metro agency with a dedicated AI operations team may not produce the same results at a smaller agency relying on sergeants to manage the implementation. The vendor's reference customer is usually the best case, not the median case.

Civil-Liberties Cost

What are the privacy, surveillance, and civil-liberties implications of this tool? This question deserves an honest answer that the agency is willing to put in writing and share with the community. It is not enough to say "we have safeguards." The question is whether the safeguards are sufficient to address the specific civil-liberties concerns the technology raises, and whether those concerns have been assessed by someone other than the vendor and the command staff.

Some tools carry civil-liberties costs that are well-characterized and manageable. AI-assisted redaction of faces and license plates from body-camera footage before public release raises a well-understood privacy concern: the risk of under-redaction. The concern is real, but it is specific, the mitigation is known (human review of the AI's output), and the alternative (not releasing footage at all, or releasing it with manual redaction that is slower and no more reliable) is not obviously better on privacy grounds. The civil-liberties cost here is manageable.

Other tools carry civil-liberties costs that are less characterized and harder to bound. A tool that continuously analyzes the behavior patterns of people in a defined geographic area, correlates that analysis with prior criminal records, and generates a real-time risk feed for dispatch has civil-liberties implications that include: the chilling effect of surveillance on protected speech and assembly; the risk of discriminatory impact if the system's training data reflects historical policing patterns that were themselves discriminatory; and the question of what happens to that data, who has access to it, and how long it is retained. These are not hypothetical concerns. The EFF and other civil-liberties organizations have documented instances of each failure mode in deployed systems. The executive considering such a tool must engage those concerns directly, not dismiss them as advocacy.

Governance Readiness

Does the agency have the policy, training, audit capability, and accountability structure to operate this tool responsibly? This is a sober question because the answer is often "partially" rather than "yes." An agency that has successfully deployed AI-assisted report writing with a verified disclosure policy has demonstrated governance capability in that domain. That same agency may not yet have the governance infrastructure to operate a real-time crime center or a drone program, not because the people are not capable, but because the policy has not been written, the community has not been consulted, the civil-liberties review has not been done, and the oversight body has not approved the parameters.

Governance readiness is also a staffing question. Who will be responsible for operating this capability? Who will audit its outputs? Who will handle exceptions, failures, and community complaints? An agency that deploys a new AI capability without assigning clear responsibility for each of those functions has created an accountability gap. When something goes wrong, and in a sufficiently large deployment something will go wrong, the question "who was responsible for catching that?" must have a clear answer. If it does not, the answer defaults to "the chief," which is the right answer in a governance sense but the wrong answer in a management sense. The chief's job is to set the standard and hold people to it, not to personally audit every output of every AI system the agency operates.

The Pre-Procurement Screen: A Practical Tool

Agencies can operationalize the framework above into a pre-procurement screen: a set of questions that any proposed AI capability must answer affirmatively before it advances to procurement. The screen is not a barrier to innovation. It is a filter that separates proposals with a defensible basis from proposals that are premature. It also generates documentation. An agency that can show that it applied a structured assessment to every AI procurement, and that the assessment produced a documented rationale, is in a significantly stronger position before a council, a prosecutor, or an oversight body than one that approved a technology because the vendor's demonstration was compelling.

The screen asks five questions. First: what specific, sourced evidence supports the value claim for this capability in a deployment context comparable to this agency? Second: what are the civil-liberties implications of this capability, and has an independent assessment of those implications been conducted? Third: what policy exists or can be written before deployment that governs the use of this capability, including prohibited uses, retention limits, oversight requirements, and disclosure obligations? Fourth: what is the CJIS compliance posture of this capability, and who within the agency is responsible for maintaining that compliance? Fifth: what is the plan for auditing the capability's outputs, handling failures, and reporting to command and to oversight bodies?

A proposal that can answer all five questions with documented, honest responses is ready to advance. A proposal that cannot is not ready for procurement, regardless of how attractive the vendor demonstration was. The pre-procurement screen is how an agency converts a discipline (responsible use-case identification) into a process (the screen) and a record (the documented responses).

The Brady/Giglio Lens on New Capabilities

Any AI capability that touches evidence, generates outputs that could be used in charging decisions, or produces records that might be reviewed in the context of a criminal case carries Brady and Giglio obligations that must be assessed before deployment. Brady v. Maryland (the 1963 Supreme Court case requiring the prosecution to disclose exculpatory evidence to the defense) and Giglio v. United States (the 1972 Supreme Court case requiring disclosure of impeachment evidence about witnesses, including officers) create constitutional disclosure obligations that attach to AI outputs just as they attach to human-generated evidence.

A new AI capability that generates risk scores, behavioral assessments, or automated recommendations about individuals creates potential Brady material: if the system flagged the defendant and that flag influenced the investigation, or if the system produced an output that was later used in a charging decision, the defense may be entitled to the system's output, the system's parameters, and information about its accuracy and validation. An agency that deploys a new AI capability without having that Brady/Giglio analysis done by a prosecutor has created a disclosure obligation it may not know about until a defense attorney asks for it in discovery.

The practical implication for use-case identification is that the legal review, including the Brady/Giglio analysis, should happen before procurement, not after. A prosecutor who sees the capability at the procurement stage and raises a Brady concern has done the agency a favor. A prosecutor who sees it for the first time in a discovery motion has identified a problem that may not be recoverable.

Building the Responsible-Innovation Culture at the Executive Level

The framework and the screen described in this lesson are tools. Tools work only when they are used, and they are used only when the organization's culture expects them to be used. Building a responsible-innovation culture at the executive level requires several things that go beyond having the right frameworks on paper.

It requires that command staff treat civil-liberties concerns as legitimate and substantive rather than as obstacles or advocacy talking points. The EFF's concerns about persistent aerial surveillance are not simply the product of an anti-police agenda. They reflect documented failure modes in deployed systems. An executive who cannot engage those concerns on the merits, who cannot explain what the agency's surveillance program does and does not do, how the data is retained, who has access to it, and what safeguards prevent misuse, has not done the governance work. Dismissing the concern is not a substitute for answering it.

It requires that there be a real answer to the question "what would stop us from deploying this?" A culture where every new capability is presumptively approved until a problem arises is not a responsible-innovation culture. It is a first-mover culture that trades long-term governance credibility for short-term capability acquisition. An executive team that can articulate, honestly, the conditions under which it would decline to deploy a new capability, even if the vendor's case is compelling, has demonstrated the governance discipline that oversight bodies and community groups are looking for.

It requires that procurement staff, legal counsel, and command staff share a common vocabulary for assessing AI proposals. The framework in this lesson provides one such vocabulary. The terminology of responsible innovation (value clarity, civil-liberties cost, governance readiness, Brady/Giglio lens, pre-procurement screen) should be standard in the agency's technology governance process, not something that gets invented fresh for each new proposal. When the vocabulary is shared, the assessment is faster, more consistent, and more defensible.

Finally, it requires that the responsible-innovation posture be communicated to the community and to oversight bodies before it is tested. An agency that presents its framework for AI use-case assessment to the civilian oversight board before a controversy arises has established a track record of proactive governance. An agency that reveals its framework for the first time in response to a community complaint has forfeited the credibility advantage that proactive disclosure provides.

Key Takeaways

• Use-case identification is an executive discipline, not a technical one. The question at the command level is not "am I using this tool correctly?" but "should we use this tool at all, and under what conditions?"
• Every proposed AI capability falls into one of three categories: high-value, lower-scrutiny (ready to deploy with appropriate governance); high-value, high-scrutiny (real value but requiring substantial governance infrastructure before deployment); or not yet (insufficient evidence, excessive civil-liberties cost, or governance requirements the agency cannot currently meet).
• The responsible-innovation line is drawn by the governance infrastructure the agency has actually built and the civil-liberties case the agency can publicly defend, not by the vendor's demonstration or the technology's theoretical capability.
• Bundled multi-year contracts approaching 45 million dollars and up to ten years in duration create relationship pressures that can distort use-case assessment. The pre-procurement screen is a tool for keeping assessment evidence-based rather than relationship-driven.
• CJIS (Criminal Justice Information Services) Security Policy obligations stay with the agency regardless of vendor contract terms. New AI capabilities require CJIS compliance review before procurement, not after deployment.
• Brady v. Maryland and Giglio v. United States obligations apply to AI outputs that touch evidence or influence charging decisions. The legal review, including the Brady/Giglio analysis, should happen before procurement, with prosecutor involvement.
• A pre-procurement screen (value evidence, civil-liberties assessment, policy existence, CJIS posture, audit plan) converts the identification discipline into a process and a record that is defensible to oversight bodies.
• A responsible-innovation culture treats civil-liberties concerns as legitimate and substantive, can articulate the conditions under which a deployment would be declined, and communicates its governance framework to the community and oversight bodies proactively rather than in response to controversy.