PoliceAI is the right idea. How does it avoid becoming a watchdog by accident?

PoliceAI, launched in June 2026, is the right idea: a national centre to develop and test AI once and give all 43 forces the approved tools. The risk is not a governance remit over how forces adopt, because it has none. It is drift by perception: a body this public, asked to comment on every incident, can come to feel like a watchdog, and that alone tempers the momentum policing has only lately started to build. Avoiding it is a matter of posture, held on purpose.

PoliceAI is three weeks old, and the idea behind it is sound. Whether it works will not be decided by its mandate but by something harder to hold onto: whether it can keep feeling like an incubator when the gravity of a high-profile public role keeps pulling it towards looking like a watchdog.

What is PoliceAI, and why was it set up?

PoliceAI launched on 10 June 2026 as a national centre for artificial intelligence in policing, and a central plank of the Police Reform White Paper published in January. The problem it answers is familiar: forty-three forces in England and Wales, each buying and testing AI tools on their own, is slow and produces wildly uneven results. PoliceAI is meant to do that work once: develop and test tools centrally, then make the approved ones available to every force.

It is backed by £75m over three years, part of a wider £140m Home Office investment in police technology (GOV.UK). The first pilots, running across up to ten forces over 2026–27, target one of the slowest parts of any investigation: triaging, disclosing and summarising digital evidence. The trials cited at launch are eye-catching: 800 hours of footage reviewed in three hours, and bulk data translated to support an organised-crime arrest. The tools are due to scale to all forces in 2027. The centre also houses a Policing AI Threat Hub, leading the national response to AI-enabled crime such as deepfake intimate-image abuse.

That is the shape of it: a body to speed adoption across forces, and to meet AI-enabled harm head-on. Both roles face outward, towards momentum.

What it is, and what it is not

What PoliceAI is, is a convenor: a place to bring forces, suppliers and ideas into the same room, set direction, and create the conditions in which adoption can happen. An environment-creator and a weather-setter.

What it is not, and what no one is asking it to be, is a body that polices how individual forces adopt AI. Policing already has bodies for assurance, inspection and complaint, and they do that work. Even the Threat Hub points outward, at criminals, not inward at forces. PoliceAI's job is to move things, not to police the movement. That distinction is easy to state now, at the start. It will be harder to hold once the centre is established and the requests begin to arrive.

Why would a watchdog posture cost policing?

Policing has historically struggled to build momentum in technology. That has started to change in recent years, but the instinct is still caution. That is understandable, given that the cost of a public mistake in policing is high. Caution is defensible in a force. It is more expensive when it sets the tone of a national centre.

The energy that moves technology forward comes disproportionately from suppliers and innovators who can work fast. If engaging with the centre feels like inviting scrutiny rather than gaining momentum, those innovators spend their effort elsewhere, in parts of the economy where they can move quicker. Nothing about this shows up as a fine or a finding. What it costs is momentum: the thing the centre exists to create, quietly redirected to somewhere easier.

How could it drift into watchdog-by-perception?

No one need decide to make PoliceAI a regulator for it to start feeling like one. The time to guard against that is now, while the centre is new. When something goes wrong, whether a tool misused or a deployment that embarrasses a force, commentators and the media will turn to the most visible national body for a response, and it may feel it has little choice but to give one. Each such request is reasonable. Answering any single one is reasonable.

The risk is in the accumulation. Answer enough of them and a centre set up to convene can begin to look, from the outside, like the place that judges. Perception is not the same as mandate, but a supplier deciding where to spend its next two years responds to how the centre feels long before it reads any terms of reference. This has not happened. PoliceAI is weeks old, and the point is that it does not have to. The posture has to be protected on purpose, from the start, precisely because the pull towards it is so reasonable at every step.

What does being on the other side of that momentum actually feel like?

I left policing in 2016, and the reason was momentum. As a detective, I had worked alongside a young technical supplier on a major statutory public inquiry, and I had seen how fast a start-up could move.

Before them, disclosure and redaction were done by hand: black pens, documents balanced on transit envelopes (anyone who served will remember them). After them, a dedicated electronic disclosure environment, able to service Rule 9 disclosure orders at a scale that had not been possible before: more than a hundred thousand documents, graded to the SECRET classification, in a day.

My mind was blown. And here is the part that matters for PoliceAI: I could see that pace from inside policing. I could not join it from there. To be part of that momentum I had to cross over. I became the supplier's first non-technical senior hire, and learned more in two years than I had expected to.

C.L.R. James, writing about cricket, asked: "What do they know of cricket who only cricket know?" The same holds for technology and policing. If you have only ever worked inside one sectoral paradigm, it is hard to see how other parts of the economy think and move, and the gap runs both ways. A national centre that wants to draw outside momentum in has to be built, on purpose, to look outward.

What keeps PoliceAI an incubator?

A few things, held deliberately against the pull of the public role.

Point outward: spend the centre's convening energy on the suppliers, researchers and other sectors that move quickly, rather than inward on the forces that lag. Keep the convening role clean of the assurance reflex: when asked to comment on a failure, route it to the bodies whose job that is, and protect the centre's own posture as the place where momentum happens. And treat pace as a measure of success, not only safety, because a body judged solely on what it stopped will, reliably, stop things.

None of this is at odds with rigour. It is about which face the centre shows the people it most needs to attract.

What to watch

PoliceAI is the right idea, and the case for a national centre is strong. The open question was never the mandate. It is posture: a choice made daily, against the pull of a public role. The bodies that set the weather are the ones that stay incubators, and staying one takes deliberate effort, because the drift runs the other way.