Editor’s note: This article is part of Police1’s “Governing AI in policing” digital report. Download the complete report here.
Artificial intelligence is often framed as a technological breakthrough. In policing, it represents something more consequential: a leadership inflection point.
Law enforcement has navigated transformational eras before. Community policing reshaped legitimacy and public partnership. [1] CompStat redefined how leaders used data to drive accountability and performance. [2] Body-worn cameras transformed transparency and evidentiary practice, forcing leaders to confront privacy, policy and public trust questions simultaneously. [3] None of these shifts were simply procurement decisions. They were command decisions that reshaped operations, supervision and public expectations.
The question facing today’s police leaders is not whether AI will influence policing. It already does. The defining challenge is whether its adoption will be governed deliberately and responsibly or allowed to diffuse incrementally without clear command direction.
The inflection point
AI represents the next structural shift in how policing processes information, analyzes evidence and supports decision-making. Like earlier transformations, it will affect operations, supervision, training, legal exposure and public trust simultaneously.
The National Institute of Standards and Technology (NIST) defines AI systems as technologies capable of performing tasks that typically require human intelligence, including perception, pattern recognition and decision support. [4] Increasingly, these capabilities are embedded within software platforms.
In practical terms, AI can accelerate report drafting, assist in video analysis, flag anomalies in large datasets and identify patterns across disparate records. It can also introduce risks of bias, opacity and overreliance if poorly governed. [5]
History shows that technologies adopted without governance frameworks create operational strain and reputational risk. The early rollout of body-worn cameras revealed the complexity of policy development, data retention and its associated costs, redaction demands and public disclosure obligations. [6] Agencies that treated cameras as simple hardware acquisitions often discovered the downstream legal and administrative consequences too late.
AI presents a similar — but broader — challenge. It does not simply document events. It shapes how information is interpreted and acted upon. That distinction elevates the issue from a technical upgrade to an expansion of command responsibility.
AI Is already inside the profession
Chiefs and sheriffs must recognize a critical reality: AI is not a future concept awaiting formal approval. It is already entering agencies through existing systems, both formally and informally. This deserves serious attention, particularly in agencies that have adopted blanket prohibitions. Banning generative AI tools such as ChatGPT does not prevent AI from being embedded within vendor platforms or being used independently by personnel.
Computer-aided dispatch and records management systems increasingly incorporate predictive text, anomaly detection and automated data sorting. Video management platforms rely heavily on machine learning for object detection and redaction. License plate recognition systems depend on algorithmic matching to identify vehicles of interest. [7] Generative AI tools are now being piloted to assist with report drafting and administrative writing. [8]
In many agencies, these capabilities arrive quietly through software updates rather than stand-alone procurements. A routine contract renewal may introduce algorithmic features that were never debated at the executive level.
At the same time, informal experimentation is also occurring, including in agencies that have issued prohibitions. Officers and supervisors may independently use publicly available generative AI tools to summarize reports, draft memoranda or assist with research. Without clear governance, however, informal adoption can outpace policy and expose agencies to legal and reputational risk.
RAND researchers have observed that law enforcement adoption of advanced analytics tools often proceeds unevenly, with varying levels of oversight and understanding across agencies. [9] In many cases, the strategic discussion follows technological integration. For command staff, that sequencing is backwards.
Essential AI terms for command staff
Artificial intelligence (AI): Systems capable of performing tasks that typically require human intelligence, such as recognizing patterns, analyzing data or generating written content. In policing, AI is often embedded within separate platforms rather than deployed as a separate tool.
Machine learning: A subset of AI in which systems learn from data to improve performance over time. Many analytics tools used in public safety rely on machine learning to identify trends or anomalies.
Generative AI: AI systems that create new content, including written reports, summaries or responses based on prompts.
Algorithm: A defined set of rules or calculations that process data to produce an output. In law enforcement, algorithms may support data matching, risk assessments or investigative analysis.
Human-in-the-loop (HITL): A governance principle requiring human judgment review of AI-generated outputs before operational or legal decisions are made.
Automation bias: The tendency to overtrust machine-generated outputs, particularly when they appear objective or technical.
Validation: The process of testing whether an AI system performs accurately, reliably and fairly under real-world conditions.
Audit trail: Documentation that records how a system was used, what data it processed and how outputs were generated.
AI governance: The oversight mechanisms and accountability structures that guide how AI systems are selected, implemented and monitored within an agency.
The real risk: Unmanaged adoption
Public discourse around AI often gravitates toward speculative extremes. For policing leaders, the more immediate risk is organizational: unmanaged adoption.
The danger is not that agencies will use AI. It is that they will use it without defined policy parameters, executive oversight and internal expertise.
The Department of Justice has warned that AI systems used in criminal justice contexts can implicate civil rights and due process concerns if not carefully evaluated. [10] NIST’s AI Risk Management Framework emphasizes governance, transparency and accountability as safeguards against harm. [11] Research further demonstrates that algorithmic systems can reproduce or amplify existing data biases if not rigorously validated. [12] When deployed in high-stakes environments, including criminal investigations, these vulnerabilities carry operational, legal and reputational consequences.
The courtroom implications are immediate. Prosecutors will face discovery questions regarding algorithmic tools used in investigations. Defense counsel will challenge reliability, validation and transparency. Agencies that lack documentation, audit trails, training records or validation studies may find themselves exposed.
None of these outcomes require dramatic failure scenarios. They stem from the absence of leadership direction. The unmanaged agency is a vulnerable agency.
Executive checklist: Is your agency leading AI adoption?
Before expanding AI adoption, command staff should be able to answer the following:
Have we identified every AI-enabled or algorithmic system in use — including vendor-embedded features?
Do written policies clearly define authorized uses, documentation standards and required human review?
Is a named executive accountable for AI governance, and is that oversight formally documented?
Have prosecutors been briefed on AI-assisted tools that may affect discovery or evidentiary review?
Do we understand training data sources, documented error rates and ongoing performance monitoring?
Have we verified how data is stored, secured, retained and who can access it within AI-enabled systems?
Have personnel received clear guidance on approved uses and the continuing primacy of professional judgment?
Can we clearly explain — in plain language — the safeguards and accountability mechanisms governing AI use?
The command staff responsibility
AI governance can no longer be treated as an IT project. It is a command responsibility. Chiefs, sheriffs and executive teams must set the tone, define guardrails and align adoption with mission and constitutional obligations. That responsibility begins with literacy.
Leaders do not need to become engineers. They do need to understand enough to evaluate vendor claims and ask informed questions:
- What data trained the system?
- What are its documented error rates?
- How is human in-the-loop review integrated?
- How are outputs preserved, documented and explained in court?
Research shows that people tend to overtrust machine-generated outputs, particularly when they appear objective. [13] Command-level literacy is therefore a safeguard against overreliance and misplaced confidence.
Guardrails must precede rollout. Policy, training and supervisory review should be established before any large-scale deployment. Governance is most effective when embedded at the procurement and design, not retrofitted after implementation, and certainly not after controversy arises. [14] Clear usage parameters, documentation standards, audit mechanisms and defined accountability structures should accompany any significant AI implementation.
Workforce preparation is equally important. Technological change without structured implementation breeds confusion, resistance and incomplete practice. Officers and supervisors need clear guidance on authorized uses, required verification steps and the continuing primacy of human judgment. Policing research consistently shows that successful innovation depends on visible leadership engagement, training investment and cultural alignment. [15]
AI will be no exception.
Leaders must also protect community trust. Policing authority rests on legitimacy, and legitimacy is shaped by perceptions of fairness and transparency. [16] AI adoption intersects directly with concerns about surveillance, privacy and bias. Agencies that deploy advanced tools without proactive communication risk undermining the very trust they seek to strengthen. [17]
Transparency does not require disclosure of sensitive investigative techniques. It does require explaining the governing principles, safeguards and accountability mechanisms behind AI use. Leaders should anticipate public questions and address them deliberately, not react defensively to them after controversy forces the conversation.
The defining question
Artificial intelligence will increasingly influence how evidence is analyzed, how leads are generated, how reports are drafted and how performance is measured.
The defining question for today’s chiefs and sheriffs is straightforward: Who sets the standards?
If leaders do not define policy boundaries, vendors will. If executives do not establish oversight structures, informal practice will.
If command staff do not build literacy, dependence will replace judgment.
AI will not determine the future of policing. Leadership will determine whether it strengthens constitutional practice, reinforces public trust and enhances professional judgment — or erodes them.
This is not simply a technology moment. It is a command moment. The question is not whether AI is coming. It is whether we are prepared to lead it.
References
- U.S. DOJ, Office of Community Oriented Policing Services. Community policing defined. Washington, DC: DOJ; 2014.
- Henry VE. Compstat management in the NYPD: Reducing crime and improving quality of life in New York City. Police Pract Res. 2002;3(3):243-259.
- White MD. Police officer body-worn cameras: Assessing the evidence. Office of Justice Programs Diagnostic Center; 2014.
- NIST. Artificial intelligence risk management framework (AI RMF 1.0). Gaithersburg, MD: NIST; 2023.
- Desmarais SL, Johnson KL, Singh JP. Performance of recidivism risk assessment instruments in U.S. correctional settings. Psychol Serv. 2016;13(3):206-222.
- Lum C, Stoltz M, Koper C, Scherer J. Research on body-worn cameras: What we know, what we need to know. Criminol Public Policy. 2019;18(1):93-118.
- Electronic Frontier Foundation. Automated license plate readers (ALPRs).
- PERF. The AI in policing paradigm: 2024 buyers guide. Washington, DC: PERF; 2024.
- Jackson BA, et al. Artificial intelligence and law enforcement: Opportunities and challenges. Santa Monica, CA: RAND Corporation; 2020.
- U.S. Department of Justice, Civil Rights Division. Artificial intelligence and civil rights. 2022.
- National Institute of Standards and Technology. Artificial intelligence risk management framework (AI RMF 1.0).
- Angwin J, et al. Machine bias. ProPublica. May 23, 2016.
- Dietvorst BJ, Simmons JP, Massey C. Algorithm aversion: People erroneously avoid algorithms after seeing them err. J Exp Psychol Gen. 2015;144(1):114-126.
- Lum C, Stoltz M, Koper C, Scherer J. Research on body-worn cameras: What we know, what we need to know. Criminol Public Policy. 2019;18(1):93-118.
- National Institute of Standards and Technology. Artificial intelligence risk management framework (AI RMF 1.0).
- Police Executive Research Forum. The police response to active shooter incidents. Washington, DC: PERF; 2014.
- Tyler TR. Why people obey the law. Princeton University Press; 2006.
