Derbyshire Constabulary says the officer allegedly used generative AI to create evidential material in multiple cases, in a probe that lands the same week the UK launched a new national PoliceAI center and told forces to stop using AI for court statements.
A single officer in Derbyshire, England has turned a question about police AI adoption into a criminal one. The unnamed officer is under investigation for allegedly using generative AI to fabricate evidential material in multiple cases, in what the Financial Times reports to be the first probe of its kind in the United Kingdom. No arrests have been made and no charges filed.
The same week, the UK formally launched PoliceAI, a new national center for coordinating AI adoption across police forces. Its interim director, Alex Murray, has already called on some forces to stop using generative AI to prepare court statements until proper checks and balances are in place, according to Futurism's reporting on the FT story.
Derbyshire Constabulary told the BBC the allegation is perverting the course of justice and that the officer has been removed from frontline duty. The Crown Prosecution Service said it is engaging with defence teams and the courts in appropriate cases, but would not comment on the live investigation.
The criminal probe reads less like a one-off lapse than as the visible crack in a larger story about how fast UK policing has adopted AI tools to write up investigations and process evidence, and how slowly courtroom-grade safeguards have caught up. The two events, the Derbyshire probe and the PoliceAI launch, are not a coincidence. They are the same story, with the criminal allegation on one side and the institutional response on the other.
That gap is what makes the Derbyshire case a structural story, not a misconduct story. It sits at the seam between two distinct failure modes that often get blurred in public debate. False arrests caused by facial recognition are an AI error problem, in which a model misidentifies a person and a human officer relies on the result. The Derbyshire allegation is a different beast: a human officer allegedly used AI to produce material that did not exist in any real-world observation, and presented it as evidence in cases that may now collapse. The first problem is a sloppy tool. The second is a deliberate misuse vehicle, and the legal exposure is correspondingly sharper. Perverting the course of justice is a serious criminal offence in England and Wales, distinct from any question of whether the underlying AI tool was safe or accurate.
UK policing has been buying evidence-synthesis AI on a commercial timetable for at least two years. Recorded Future News has documented the rise of police-focused platforms including Tranquility AI's TimePilot, Allometric's AirJustice, and Truleo, all of which pitch faster report-writing and case summarisation to overstretched departments. The same reporting surfaces the defence-side critique. Center for Democracy and Technology counsel Tom Bowman has warned that generative tools hallucinate, omit context, and miss exculpatory evidence, and that the courtroom is the wrong place to discover that.
Lighter-touch precedents are already on the file. A Utah police report generated with AI help hallucinated that an officer "turned into a frog," per Futurism's coverage. In Maine, officers were caught posting AI-tampered "drug bust" photos online. Both episodes were embarrassing. The Derbyshire case, if the allegation is substantiated, is the first one in the UK in which a force has responded by treating an officer's use of generative AI as a criminal matter, not a policy hiccup.
That distinction will shape what comes next. The Crown Prosecution Service's involvement, alongside Derbyshire Constabulary's professional standards department, suggests any prosecution would run through the ordinary criminal courts rather than a misconduct panel, which raises the bar for evidence and gives defence teams a clean forum to challenge the integrity of the wider case file. Lawyers who handle disclosure fights in England and Wales are already watching for the moment when a defendant asks, on oath, whether any document in the prosecution bundle was drafted or altered by a generative model.
PoliceAI's guidance to forces to pause their use of generative AI for court-bound paperwork, issued the same week the Derbyshire news broke, is the institutional response the situation had been asking for. It is also a tacit admission that adoption has been racing ahead of the controls a courtroom requires. The two stories belong together: an officer under criminal investigation, and a national center that has just told forces, in writing, that they are not yet ready to use this technology for the work that ends up in front of a judge.