Police AI director: deploying AI into UK policing is ‘building the train on its journey’

TL;DR:

• Alex Murray, interim director of Police AI within the College of Policing, said the function’s role is to “deliver responsible AI into the hands of frontline staff” — not to run pilots, but to scale tools that have been tested and evaluated.
• Plans are in train for a public-facing registry of AI systems used in UK policing and an AI laboratory to independently test supplier claims before national rollout.
• Murray framed responsible AI in policing as anchored in legitimacy and public consent — accepting slower deployment in exchange for trust — with control rooms, disclosure processes and digital-evidence review the practical near-term areas.

Murray’s positioning matters because it locks in a specific operating model for one of the most sensitive UK AI deployments. Police AI as a national function is explicitly being set up not to chase the proof-of-concept demonstrations that have characterised much UK public-sector AI work to date, but to do the harder structural job of scaling what works and weeding out what does not. The “building the train whilst it’s on its journey” framing acknowledges the timing pressure but does not concede the verification standard.

A registry-and-lab model goes further than other UK departments

The plan for a public-facing registry of policing AI systems is the most striking single commitment. UK government has talked about algorithmic transparency for several years — the Algorithmic Transparency Recording Standard has existed in some form since 2022 — but uptake has been patchy. A registry covering live police AI, paired with an independent lab testing supplier claims before national procurement, would create a stronger empirical basis for accountability than most other UK public-sector AI rollouts currently have. The same week, the Bar Standards Board completed UK legal regulators’ coverage of barrister AI use, Defra confirmed AI pilots in policy drafting, and Derby Council’s three-year AI customer-service experience showed a 50% resolution-rate plateau — collectively a UK-wide picture of public-sector AI moving from pilots into accountable deployment.

Murray’s career trajectory — West Midlands Police, the Met, West Mercia, the National Crime Agency, and the NPCC AI lead role — gives the function institutional credibility within policing that previous national-tech programmes have sometimes lacked. The application areas he flags — disclosure, digital-evidence review, control-room triage — are the ones with the strongest current-demand cases and the clearest accountability constraints. Disclosure in particular has been the subject of significant criticism from defence lawyers and judges in recent years; how AI changes the search-and-triage step matters for the next miscarriage-of-justice cycle as well as the cost equation.

Looking forward

Watch for the registry’s first publication and the lab’s first published evaluations — both will be the practical tests of whether the responsible-AI framing survives delivery pressure. For UK SMEs selling AI into policing, the implication is that supplier-claim testing is moving from buyer-controlled to lab-independent, which raises the bar on evidence quality. For UK firms thinking about AI governance more generally, the policing model — registry plus independent lab plus stated public-consent threshold — is a more concrete template than the more abstract “responsible AI” framings most enterprises have adopted to date.