Artificial intelligence is being adopted across policing and the wider criminal justice system of England and Wales faster than the rules designed to govern it, according to major new research published by Northumbria University.

The study—the most comprehensive mapping exercise of its kind to date—identifies 70 AI tools already deployed, trialed or under development across the criminal justice system, from contact-center triage and witness statement drafting to digital forensics and facial recognition.

Led by Northumbria University in partnership with the Universities of Glasgow, Northampton, Leicester, Newcastle and Cambridge, the four-year PROBabLE Futures project was funded by Responsible AI UK as a 4.2 million-pound Keystone Project. The team spent the past year building an interactive mapping tool and conducting interviews between May 2025 and January 2026 with police officers, government officials, legal professionals, oversight bodies, academics and technology vendors.

Of the 70 tools identified, 27 are already live, with around 34 at the trial or pilot stage. More than half, 52%, come from commercial vendors, with most activity concentrated at the community policing, intelligence and investigation stages of the criminal process.

The research delivers a clear central finding: AI is already generating real value in transcription, redaction, crime analysis, vulnerability identification and officer welfare—but only where it has been carefully designed, matched to clearly defined operational problems and robustly evaluated. Across the wider system, adoption is outpacing the institutional, evidential and governance frameworks needed to support it.

A particular focus of the report is the concept of the "human in the loop"—the principle that a person should review and remain accountable for AI outputs. The research finds this principle often exists in name only, providing a false sense of assurance rather than a genuine safeguard.

The report also highlights a counterintuitive risk: As a tool approaches near-perfect accuracy, people may stop checking its outputs, meaning rare errors are more likely to go undetected and carry serious consequences. This danger grows when AI systems are linked together in sequence, with each stage potentially inheriting and compounding errors introduced earlier.