← All stories

Leadership & Policy

Police AI adoption raises a governance question: who is accountable?

A November 19 discussion revisits an August NPI report on AI readiness, highlighting the gap between deploying tools and defining training, review and responsibility.

OfficerList editorial headline graphic about police AI governance; not an incident photograph.
OfficerList editorial headline graphic about police AI governance; not an incident photograph.

A November 19 Police1 discussion with National Policing Institute President Jim Burch puts a practical question before police leaders: who is responsible for managing artificial intelligence after an agency starts using it? The discussion revisits NPI's report on the gap between acquiring AI tools and preparing an organization to oversee them.

The underlying report is not a new November study. NPI announced it on August 11, 2026, drawing on an April roundtable in San Francisco hosted by Microsoft. The fresh development is the Police1 discussion of that work, not a new nationwide count of police AI use.

That distinction matters because the report's headline numbers describe participating agencies, rather than American law enforcement as a whole. They offer a starting point for questions about readiness, not a national adoption rate.

What the roundtable found

According to NPI's announcement, the gathering brought together chiefs, sheriffs, senior command staff, analysts and technology strategy officers from more than a dozen agencies. Participants represented organizations ranging from large urban departments to small-city agencies and regional sheriff's offices.

NPI reported that 83% of participating agencies had formally deployed at least one AI tool. It also said 44% had provided no AI-specific training to personnel, while 61% had designated a person or committee to oversee AI governance. Those figures should be read within the limits of the roundtable, which NPI describes as a structured listening session conducted under Chatham House rules.

The announcement does not establish a representative national sample. Nor does appointing a committee demonstrate that an agency has resolved every operational issue. NPI specifically identifies questions about approval of individual uses, audit trails and responsibility when an AI output is wrong.

For officers and supervisors, those questions connect policy to everyday work. Knowing that a tool is available does not, by itself, explain which assignments it is approved for or how its output should be checked.

Readiness is more than procurement

NPI's argument is organizational: data infrastructure, governance, workforce knowledge and community trust need attention alongside the software. Its announcement calls for a strategy for preparing agencies, rather than treating the evaluation of individual products as the whole task.

The Police1 episode page similarly distinguishes purchasing technology from being ready to manage it. It describes the discussion as a call for a national, field-led strategy, not a federal mandate. Neither source announces a new legal requirement that departments must meet.

The topic also extends beyond a department's formal purchase list. NPI's announcement raises the issue of officers and deputies using AI through personal devices and accounts, potentially without agency approval. That is a concern reported in the roundtable account; the sources do not provide a nationwide estimate of unauthorized use or establish misconduct by any particular officer.

Police1's episode page lists examples and topics explored in the conversation. This article relies on that written page and NPI's announcement, not on an independently reviewed transcript of the full interview. It does not treat advertised discussion points as tested operational results.

OfficerList analysis: make accountability specific

An agency reviewing its AI readiness could begin by listing approved tools, their intended uses and the person responsible for each. The useful distinction is between permission to access an application and permission to use it for a particular police task. These are OfficerList implementation questions, not additional findings from the roundtable.

A second review could follow an output through the workflow. Who checks it before it enters an official record? What happens when it conflicts with source material? How is a correction documented? Answering those questions makes responsibility more concrete than assigning a general technology lead.

Training can then be tied to actual duties. A supervisor reviewing AI-assisted work may need different preparation from the employee producing it. A policy review can also ask how personnel should handle information in unapproved services and where they should take an uncertain use case before proceeding.

Community communication deserves the same specificity. Explaining what an agency proposes to do, what remains a human decision and how concerns can be raised is more informative than describing a product simply as innovative. Whether those practices improve trust would require evidence beyond these sources.

What the evidence does not show

The reviewed material does not demonstrate that AI reduces crime, improves report accuracy or saves a specified number of staff hours. It also does not prove that one governance model works for every department. NPI is advocating organizational preparation based on a listening session, and the November discussion brings that argument back into view.

The immediate takeaway is a management question rather than a purchasing recommendation: can an agency explain who approves, checks and takes responsibility for the AI-assisted work it already allows?

Reporting note: Based on the Police1 episode's written description and NPI's August 11 announcement. OfficerList conducted no independent interviews. The full report and podcast were not independently reviewed; practical recommendations above are labeled analysis.

Sources and reporting note

This summary is based on the reporting linked below. OfficerList did not independently investigate the incident.