The challenge of delivering AI into UK policing

The role of AI in policing is often framed in terms of future risk, opportunity and transformation. But in the latest episode of our For Every Response podcast, Catherine Levin sat down with the interim Director of Police AI, Alex Murray, to explore what it actually means to deliver AI into frontline policing today, and why the challenge sits as much in delivery and structure as it does in technology.

Murray has spent his career across UK law enforcement, including roles in the West Midlands Police, Metropolitan Police Service, West Mercia Police and the National Crime Agency. He also served as the National Police Chiefs’ Council lead for AI before taking on his current role as interim director of Police AI, within the College of Policing.

The thread running through his career has always been a focus on improvement and evidence-led policing.

“The police service needs to be at the frontier of development. I’m profoundly interested in evidence based practice, which is a real driver for innovation, and so much of that is now in the technology space.”

That perspective now shapes a national function that is focused not on experimentation alone, but also on scaling tools that are proven to work.

Murray explains to Catherine where Police AI sits in the wider ecosystem. “We’re squarely in the face of delivery, that our job is to deliver responsible AI into the hands of frontline staff.” And just as clearly, he defines where it does not sit.

“We’re not in the proof of concept space, where a lot of people operate, but we’re in the space of understanding what works, testing and evaluating it, scaling it and engineering it into police forces around the country.”

That distinction matters, he argues, because the expectation on policing is to move beyond discussion and into implementation, even as the technology continues to evolve rapidly. It is a reality he describes as fundamentally different from traditional programmes of work in policing and public sector delivery: “It’s like building the train whilst it’s on its journey.”

A core theme running through the conversation is what “responsible AI” actually means in a policing context. Murray sees it as being anchored in legitimacy and public trust. “We survive on public consent,” he tells Catherine.

That requirement, he says, shapes how AI can be deployed, particularly where it may influence decision-making or evidence handling, but it also slows down implementation by design.

“We want to deliver responsible AI. That means you’ve got to do it a bit slower, but it means that the public can be reassured that we’re not just throwing AI out willy-nilly.”

Much of the practical application discussed in the episode sits in areas where policing already faces significant demand and complexity. From disclosure processes to digital evidence review, AI is being positioned as a tool to reduce administrative burden and improve accuracy, but Murray is careful to stress that deployment is not about removing human judgement. The same applies in control rooms and contact centres, where AI is already supporting call handling and triage in some forces, including within the West Midlands Police.

A significant part of the programme’s work is focused on ensuring tools are properly tested, evaluated and publicly accountable before being scaled. That includes plans for a registry of AI systems used across policing.

“We want a public facing registry of pertinent AI that people can go on and see that we’re using AI in this context, and can be assured that it’s tested and evaluated.”

Alongside this, an AI laboratory function is being developed to independently test supplier claims before tools are rolled out nationally.

As policing continues to modernise alongside wider reform, Murray argues that the success of AI will depend on whether it can be delivered at scale without losing accountability or public trust. He is, above all, keen that AI in policing is not positioned as a replacement for human judgement, but as a shift in how time, capacity and attention are used across the system.

Catch up with the episode below.