by Professor Marion Oswald, MBE
This is a version of a talk given by Professor Oswald at the APCC/NPCC Summit (Radical Futures breakout) on 17 November 2026
A few weeks ago, I read an article in a policing-focused publication that had been paid for by an international technology company headquartered outside the UK. Linked with the launch of Police AI, the article argued that slow governance causes harm because of delays in getting commercial products into the hands of police officers. More positively, the article argued that commercial providers should develop their products responsibly to make it easier for the public sector to approve them. In relation to their particular product, they asserted that nothing is released until an officer has reviewed what the system has produced and approved it. Attention was also drawn to the company’s own advisory council and the company-commissioned studies of racial bias and quality.
It’s a positive step when commercial providers proactively provide transparency information about their product development and performance, and provider evaluations can provide one bit of the picture. Police forces cannot however sit back and just rely only on that as the sole evaluation. Let’s think of an analogy of a Russian doll: a set of wooden hollow dolls one inside another. The solid doll right in the middle is the technology, the AI product, say. The commercial provider can tell you all about that: how it was built, the results from testing in the lab or perhaps in some specific operational environments, and some measures of bias. But there are other contextual layers that you must consider – building your Russian doll – to work out whether what you are planning to do with your technology is legitimate, will maintain public confidence, and actually lead to the outcome you’re after.

Your next layer is the policing context: what are you trying to achieve? What messy police data or operational context will you encounter? How does this impact performance? Is the deployment of the technology different to that for which it was designed and evaluated? How will officers respond? There is much reliance, over reliance some would say, on the human-in-the-loop to check AI outputs. But as the recent HMICFRS report on retrospective facial recognition demonstrates, this is risky. HMICFRS found that some investigators relied too heavily on RFR without treating it as intelligence, leading to inappropriate investigative action. This itself could lead to legal issues under the Police and Criminal Evidence Act and evidential frameworks. And of course, in England & Wales alone there are 43 slightly different policing contexts to choose from – each legitimate but not always the same.
Next comes policing by consent, something we hear a lot about, but perhaps is not always treated as seriously or thoughtfully as it should be when it comes to technology. There are many ways in which trust in the police can be undermined, including a lack of prioritisation of local antisocial behaviour or violence against women and girls, and crimes committed by police officers themselves. In relation to technology, the recent Home Office public dialogue research on AI and policing underlined that the public want to see that AI improved policing outcomes, including the victim experience. But they were also concerned about fairness and integrity, so were not prepared to give the police free rein.
The HMICFRS report on RFR emphasises transparency and public consultation. Furthermore, the government’s proposed new legal framework for biometrics and similar technology in law enforcement is to include a statutory requirement for public participation in oversight processes. This is going to require new thinking by those in leadership positions. Relying on a commercial provider’s own advisory board is not going to cut it.
Some forces, such as West Midlands Police (I chair the WMPCC and WMP Technology and Operational Ethics Panel) already have mechanisms that facilitate such participation. The Home Office Science and Technology Ethics Advisory Committee gives advice on national police technology rollouts and strategic data issues, providing another route for obtaining independent advice on projects raising national issues. Police forces should start to consider the options for public and independent involvement so that they can benefit from the valuable insights that independent engagement brings.
Working with an independent advisory body should not be regarded as a tick-box exercise or as something that will cause problematic delays. It’s your way of getting what academics would think of as a ‘peer review’. It might make us a little annoyed to begin with, but when we sit back and review what’s been said, we realise that taking the advice into account will make our work much better. And for the police, ensuring you understand the community implications of your proposed technology will be vital for success. It helps gives reality to the Peelian tradition that the police are the public and the public are the police.
But there’s a final layer of our Russian doll that we must consider. This is the criminal justice outcome. Police are not in control of that, but what they do impacts on it. My project PROBabLE Futures has recently conducted a major mock trial in Glasgow alongside the Procurator Fiscal’s office and Police Scotland. It was evident from this exercise that if AI-based evidence is introduced into a trial when it is not fully understood, accurate or trusted, it can be more of a hindrance than a help.
To conclude, to achieve better innovation for the long term, we need to consider all the layers of the Russian doll, not just the technology one in the middle. As we’ve seen from the Government’s announcement about the new biometrics legal framework, independent regulation and oversight in this area is coming, and will bring the police a little bit closer to the existing Investigatory Powers Act regime already in place for the intelligence agencies. Also on its way is the requirement for more structured and embedded engagement with independent experts and community representatives, something to plan for now. You don’t need to start from scratch – there’s best practice out there already. Trust the communities outside policing to have something useful to tell you about technology implementation.
Join us as an official partner if your organisation is interested in Responsible AI research, innovation and skills. We are looking for partners including businesses, funders, charities, creative organisations, industry associations, think tanks and more. Find out more and sign up
RAi UK will store your data in accordance with the General Data Protection Regulation 2017 (GDPR). We will not share your data with any third parties and you are given opportunities to unsubscribe at any time within the electronic communications you receive, or by emailing info@rai.ac.uk