Police application of generative artificial intelligence: A scoping review

Generative artificial intelligence (GenAI) has rapidly entered policing and law enforcement discourse, yet the evidence base concerning what has been tested and what outcomes have been reported remains absent. This study presents a scoping review of academic literature on the actual or theoretical application of GenAI in policing and law enforcement since the emergence of Transformer-based architectures in 2017. In total, 42 studies are examined and data was charted by function, policing context, GenAI model type, and methodological design. Results indicate a literature base dominated by benchmarking, simulation, and proxy evaluations, with only a small number of operational or practice-proximate comparative studies. The article analyses the studies thematically using ‘what works, what does not, and what appears promising’ framing, and finds that most evidence is best characterised as promising rather than conclusive. More consistent early capability signals appear in bounded, structured, and codifiable tasks such as information extraction, triage, and thematic coding, although these findings remain largely derived from benchmark, simulation, or proxy evaluations. In contrast, cognitively complex functions remain inconsistent and risk prone. The review highlights major gaps in independent evaluation, operational testing, geographic coverage, and governance assessment, and proposes a research agenda to strengthen the literature available.