NextFin News - A Derbyshire police officer is under criminal investigation over allegations that AI systems were used to create evidential material in a number of cases, in what appears to be the first known case of its kind in the UK. Derbyshire police said the officer had been removed from frontline duties, no arrests had been made, and the investigation into alleged perverting the course of justice remained at an early stage.
On the surface this looks like a misconduct case; the real issue is whether AI can be allowed anywhere near the evidential chain without breaking it. This is not about whether software can draft text faster — it is about whether a public authority can defend every line of a statement, summary or exhibit in court. If it cannot, the business case for using AI in that process collapses because the cost is no longer time saved but prosecutions weakened, cases delayed and credibility damaged.
Alex Murray, who heads the National Police Chiefs’ Council’s Police AI centre, has already warned a number of police forces to stop using AI systems to prepare court statements and other tasks because the tools might not be reliable enough. That warning matters less as a technology critique than as a risk judgment from the part of policing charged with keeping procedure intact. The beneficiaries of that caution are defendants, prosecutors and forces that want cases to survive scrutiny; the pressure falls on vendors and police managers who treated generative tools as an extension of office software rather than a source of legal exposure.
The logic holds up because criminal justice has a harsher standard than most workplaces. In an ordinary enterprise setting, a flawed AI output can be edited, logged and replaced with limited fallout. In criminal justice, every break in provenance creates a point of attack: who wrote the text, what source it relied on, what was changed, and whether any fabricated or distorted detail entered the record. The real trade-off is speed versus admissibility. Plenty of AI tools, from transcription to document triage to pattern recognition, may still have a place if tightly controlled, but once AI helps produce evidence the threshold moves from convenience to verifiability, and the math doesn't add up yet.
What still needs to be verified is straightforward and consequential: what kind of evidential material was created, how many cases were affected, whether supervisors knew, and whether audit trails exist that can show exactly what the systems did. This is not yet a verdict on AI in policing, and it is not proof that all AI use in law enforcement is compromised. The risk nobody is talking about is that this turns out to be less a rogue-use story than a governance failure spread across forces that adopted tools faster than they built controls. Derbyshire police says the investigation is continuing, the officer remains off frontline duties and no arrests have been made. Other forces are already being told to slow down or suspend certain AI uses before those systems reach court statements and evidence files.
Explore more exclusive insights at nextfin.ai.
