Large Language Model Technologies in Policing: Early Lessons in Artificial Intelligence from Research and Practice

Abstract

Artificial intelligence (AI) is increasingly framed as the next technological “revolution” in policing, promising faster report writing, automated transcription, and improved accountability. This chapter evaluates that claim by focusing on large language models (LLMs) and related language-processing tools used for police documentation and body-worn camera (BWC) review. Across randomized trials, agency pilots, and evidence from adjacent professional domains, findings converge on a consistent pattern: AI-assisted report writing rarely produces measurable time savings. Instead, LLMs shift work from drafting to verification and correction, offsetting speed gained during initial text generation. This dynamic helps explain a central adoption puzzle: why departments continue to adopt tools that do not reduce workload. AI’s influence is better understood across two dimensions of value: efficiency (time and workload) and behavior and professionalism (how officers communicate and how work is documented and reviewed). While efficiency gains are scarce, emerging evidence suggests AI can improve behavior and professionalism when embedded in supervisory routines, including transcript-based feedback systems that shape officer communication and reduce substandard professionalism. AI drafting tools may also improve documentation clarity and accessibility for officers who struggle with writing. Rather than transforming policing through administrative acceleration, AI appears more likely to reshape documentation practices, supervisory review, and feedback structures. Four structural constraints shape implementation: workflow disruption, legal and evidentiary complications, uneven digital literacy, and underdeveloped governance. Whether AI produces meaningful value depends on how agencies address these conditions through transparency, institutional readiness, and rigorous evaluation.