Artificial intelligence (AI) is no longer an emerging issue for law enforcement. It is already here, evolving daily, and influencing how public safety agencies operate.
Recently, I read an article about research underway at East Texas A&M University, where researchers are studying whether AI-generated images can help law enforcement create fairer eyewitness lineups. The project seeks to determine whether AI-generated “filler” photos could reduce bias in identification procedures or unintentionally create new problems that affect witness decision-making. The fact that researchers are exploring AI’s role in something as consequential as eyewitness identification demonstrates how rapidly this technology is moving beyond administrative functions and into core investigative processes that may affect constitutional rights, criminal prosecutions, and public trust (Wright, 2026).
Many police leaders still view AI as a future issue or believe it only applies to agencies that have formally purchased AI products. That assumption is becoming increasingly risky.
If an officer uses a public AI platform to summarize a report, draft correspondence, analyze information, translate a statement, organize evidence, or generate investigative leads, the agency is already dealing with AI-related risk. The question is not whether AI will affect policing. The question is whether law enforcement leaders will establish governance and controls before an avoidable problem forces action.
The Gray Rhino Is Already Charging
In The Gray Rhino: How to Recognize and Act on the Obvious Dangers We Ignore, Michele Wucker (2016) describes a “gray rhino” as a highly probable, high-impact risk that is visible to everyone yet often ignored until it becomes a crisis. Unlike an unexpected event, a gray rhino provides ample warning. Leaders simply fail to act.
AI fits this definition perfectly.
Policing research has identified expanding AI applications in criminal investigations, intelligence analysis, public communication, workforce management, report generation, and administrative functions. Researchers have consistently emphasized the need for governance structures addressing legality, accountability, transparency, security, ethics, and community trust (Halford, 2025; Sorell, 2024).
Waiting until after a controversy, courtroom challenge, data breach, or civil lawsuit occurs is not a strategy. It is simply a delayed reaction.
“We Don’t Use AI” Is Not a Risk Management Plan
One of the greatest misconceptions facing law enforcement today is the belief that agencies can avoid AI risk simply by deciding not to purchase AI tools.
The reality is that many employees already have free access to powerful AI platforms on their desktops, laptops, and smartphones. Without clear guidance, officers and professional staff may unknowingly enter sensitive information into systems the agency has never vetted.
The risk is not limited to vendors. It also involves employee behavior, data governance, and organizational oversight.
Research examining machine learning in policing warns that leaders must understand not only what a system produces, but also how it reaches its conclusions. Police executives and policymakers should critically evaluate the data being used, the outcomes being measured, and the assumptions embedded in AI-supported decision-making (Vestby & Vestby, 2021).
Without policy direction, agencies effectively surrender control over how AI is being used.
The Technology Is Powerful, but It Is Not Always Right
Generative AI can produce reports, summaries, analyses, and recommendations that appear polished, professional, and authoritative. However, researchers have documented that AI systems can generate inaccurate, unsupported, or entirely fabricated information, often called “hallucinations” (Dang et al., 2025).
Equally concerning is the risk of automation bias. Research has shown that people can place excessive trust in automated recommendations, even when those recommendations are wrong. While these studies were conducted in other professional disciplines, the findings are highly relevant to policing, where employees may be tempted to accept AI-generated information simply because it appears credible (Kücking et al., 2024).
An AI-generated report does not become accurate simply because it sounds professional.
The human user must always verify facts, assess credibility, establish probable cause, and exercise professional judgment.
The Legal Risks Are Real
The legal exposure associated with AI will likely arise from ordinary operational decisions rather than science-fiction scenarios.
Questions that attorneys, courts, prosecutors, and oversight bodies may eventually ask include:
- Who authorized the use of the AI tool?
- What information was entered?
- Was confidential, CJIS-protected, medical, personnel, or investigative information disclosed?
- Did AI influence a probable cause determination?
- Was the output independently verified?
- Can the agency explain and reproduce the process?
- Was the use documented and auditable?
Courts and juries may be less interested in what technology was used than whether agency leadership exercised reasonable oversight.
The absence of a policy will not eliminate liability. It may make liability much harder to defend.
Every Agency Needs an AI Policy
Whether an agency actively deploys AI systems or not, leaders must establish clear expectations.
At a minimum, agency policies should:
- Define what constitutes AI.
- Require authorization for official AI use.
- Restrict the entry of protected information into unauthorized systems.
- Require independent verification of all AI-generated content.
- Preserve meaningful human review and accountability.
- Establish documentation and audit requirements.
- Identify prohibited uses.
- Address procurement and vendor oversight.
- Require employee training.
- Mandate periodic policy review.
These principles align with emerging research that calls for structured governance, ethical review, stakeholder involvement, and ongoing evaluation of AI systems used in policing (Halford, 2025; Sorell, 2024; Vestby & Vestby, 2021).
Leadership Means Acting Before the Crisis
Law enforcement has a long history of adapting to new technologies. We are seeing this now with Automated License Plate Reader (ALPR) technology, particularly systems such as Flock Safety, where technological capabilities have advanced more quickly than public understanding, legal analysis, and policy development. Many agencies have found themselves developing retention standards, audit requirements, transparency measures, and governance frameworks after the technology had already been deployed. The lesson is not that the technology is inherently problematic, but that innovation often outpaces policy. AI will be no different (Brayne, 2017; Sorell, 2024).
The goal is not to ban innovation. AI offers legitimate opportunities to improve efficiency, reduce administrative burdens, and support decision-making. However, history teaches us that new technologies often create risks before policies catch up.
The agencies that will succeed are not necessarily those that adopt AI first. They will be the agencies that adopt it responsibly.
Wucker’s (2016) Gray Rhino framework reminds us that leadership failures often occur not because threats were hidden, but because they were obvious and ignored. AI is an obvious and rapidly approaching challenge for law enforcement. We know it is here. We know our employees can access it. We know it will become increasingly integrated into public safety operations.
The question is whether we will develop the policies, oversight mechanisms, and accountability structures now, or wait until a preventable problem forces us to act.
The gray rhino is in plain sight.
The time to prepare is now.
–Chief Jeff Caponera serves as Chief of Police in Grafton, Wisconsin, bringing more than 30 years of law enforcement and executive leadership experience. Throughout his career, he has focused on building professional, accountable organizations grounded in service, trust, and doing the right thing, especially when it is not the easiest choice.
_________________________________________________________
References
Brayne, S. (2017). Big data surveillance: The case of policing. American Sociological Review, 82(5), 977-1008.
Dang, A.-H., Tran, V., & Nguyen, L.-M. (2025). Survey and analysis of hallucinations in large language models: Attribution to prompting strategies or model behavior. Frontiers in Artificial Intelligence, 8.
Halford, E. (2025). The Transformer Led Policing model: A framework for applying generative artificial intelligence in policing. Policing: A Journal of Policy and Practice, 19, paaf027.
Kücking, F., Hübner, U., Przysucha, M., Hannemann, N., Kutza, J.-O., Moelleken, M., Erfurt-Berge, C., Dissemond, J., Babitsch, B., & Busch, D. (2024). Automation bias in AI-decision support: Results from an empirical study. Studies in Health Technology and Informatics, 317, 298-304.
Sorell, T. (2024). AI-related data ethics oversight in UK policing. Policing: A Journal of Policy and Practice, 18, paae016.
Vestby, A., & Vestby, J. (2021). Machine learning and the police: Asking the right questions. Policing: A Journal of Policy and Practice, 15(1), 44-58.
Wucker, M. (2016). The Gray Rhino: How to Recognize and Act on the Obvious Dangers We Ignore. St. Martin’s Press.
Wright, T. (2026, September 8). Can AI make police lineups fairer? ETAMU researchers aim to find out. MyParisTexas.
Reprint Disclaimer: This article is reprinted with permission. Reprint and publication rights were expressly granted by the author for publication on Strategic AI for Law Enforcement.
