Police use of AI: constrained and cautious or out of control?

In June 2026, Metropolitan Police Commissioner Mark Rowley gave a speech arguing that police forces should not be expected to wait for legislation every time they want to use a new technology.

But for years, police forces have introduced all kinds of AI systems without any governance or oversight, from live facial recognition to predictive analytics. Probable Futures just released a map that found over 70 AI tools in use across England and Wales.

The Commissioner is right; it would be impractical to wait for Parliament to create fresh legislative frameworks every time the police want to do something new. That’s why the UK needs: an overarching legislative framework that can adapt to the pace and risks of emerging tech; rules that ensure the police are procuring rights-respecting tech; and an empowered regulator that can hold police to account.

Three case studies underscore the importance of getting this right and why AI’s use in our criminal justice system must be carefully governed.

Predictive policing in secret

Liberty Investigates’ reporting has found over 90% of Bristol’s population had their data fed into a predictive database used to build machine learning models.

Created by Bristol City Council and Avon and Somerset Police, these models assigned risks scores to adults and children and create a ‘picture of threat, harm, and risk in the region’. The data included mental health records, housing status, teenage pregnancies, and free school meals. Two of the risk-scoring models were quietly shut down after years of use, when the scores were found to be ‘inaccurate’ and ‘not fit for operational use’.

None of this was disclosed to the people of Bristol, who were kept in the dark despite a 2016 Police Ethics Committee warning that ‘the public must be informed as to why and how you are carrying out such [predictive] processes.’ Avon and Somerset Police said they have a dedicated ethics group to review their AI systems, but a 2023 records request showed no meeting of the group had been held because ‘no model has been produced for which potential ethical issues have been identified.’

Hallucinating intelligence

In 2025, West Midlands Police came under fire after banning fans of Maccabi Tel Aviv football club from a match in Birmingham. Whatever the merits of that decision, our concern centres on how it was reached.

This ban was based on intelligence hallucinated by Microsoft’s Copilot, which was not properly fact-checked by police officers. Copilot included inaccurate information about violence that occurred at a match in Amsterdam. A Home Affairs inquiry found the ban caused ‘serious damage to trust’ in police in the West Midlands.

Biasing justice

Several rape convictions are currently under review because Derbyshire detective prepared evidence by giving ‘biased’ prompts to an AI chatbot. The officer allegedly told the system to maximise the impact of the victims’ impact statements, and prepared briefings for prosecutors by prompting the software to ensure they would authorise charges.

What can we learn from this?

• Voluntary transparency and oversight mechanisms are failing – police forces cannot be allowed to continue marking their own homework. In 2023, all police chiefs in England and Wales signed up to a Covenant on AI which bound them to ensuring their use of AI is transparent, accountable, and responsible. We are still waiting on them to keep this promise.
• Accountability cannot be outsourced to an algorithm – AI systems will produce errors, but cannot be held accountable for them. People must be able to understand how decisions are made and by whom, so they can challenge them. Ultimately, decisions must rest with a human and so must liability.
• The harm is already here – police officers are adopting this tech in a risky way, without proper training, and the harms of that approach are already impacting individuals and communities.

Thankfully, the Government has just created Police.AI, a new body which will have the expertise to scrutinise police AI systems. However, to be effective, it needs to have power.

If the Government is serious about keeping people safe and ensuring that AI doesn’t rip through the criminal justice system’s careful accountability mechanisms, then it needs to empower Police.AI to regulate the two things in its name. The Government should also attach conditions to technology funding: any force seeking money for new tools must show the technology is effective, safe, and rights compliant.

Who will watch the watchers?

To quote the Met Commissioner, ‘Criminals don’t have to play by any rules. They do not care about protecting your data, they actively seek to exploit it. They are not constrained by governance, or procurement rules.’

But to build public trust around new technologies, it is essential that police do follow the rules.

The police have the power to spy on us, to use force against us, to take away our liberty if we are suspected of a crime. If we do not place rules around the powers police have, and the technology they increasingly use to exercise that power over us, it is bad news for the communities who have good reasons to distrust the police and bad news for our democracy.

Technology reflects trust – it doesn’t create it. Police forces that haven’t earned public confidence through transparency and accountability won’t find it in AI. They’ll just automate distrust.