Criminals could walk free without controls on AI use, police watchdog warns

Criminals in England and Wales could wrongly walk free without proper policies to stop misuse of AI in investigations, the head of the police watchdog has warned, as she urged forces to boost safeguards against the “tempting tool”.

Rachel Watson, director-general of the Independent Office for Police Conduct, also told the FT that officers would be at unnecessary risk of being disciplined for misconduct unless chiefs set clear standards on avoiding unauthorised use of AI software.

“It could lead to a mistrial,” Watson said of AI misuse in an interview. “There’s always the potential to lead to someone not being convicted because of incorrect information inserting itself in.”

The warnings by Watson came after what she said was the “jolt” of shock that had run through the policing system after the FT revealed that a detective constable in Derbyshire Police was under criminal investigation for allegedly using AI to “create evidential material”.

The officer, who has not been publicly named, has been suspended from frontline duties while the force investigates him for potentially conspiring to pervert the course of justice by using the fast-evolving technology to prepare court documents.

People familiar with the case have said the officer instructed a chatbot to prepare documents that would maximise the chance of conviction and a severe sentence.

News of the incident highlighted the risks to policing from AI shortly after the launch of PoliceAI, a specialist unit intended to help forces to generate efficiency savings through using new software.

Watson, formerly policing director at the Home Office, said she had also written to the National Police Chiefs’ Council to warn the representative body about the risk to police integrity from the technology’s tendency to make up facts.

The IOPC has received only four referrals over AI misuse cases and at present has two live investigations into AI issues. One is a probe of a Derbyshire Police detective chief inspector who allegedly used an AI system to maintain legally critical logs of the decisions that she took during investigations.

The other is over how West Midlands Police came to use a dossier of evidence including facts hallucinated by Microsoft’s Copilot AI system to recommend banning visiting fans of Israel’s Maccabi Tel Aviv from attending a match against Aston Villa last year — a decision that provoked a political outcry.

Watson said forces needed to refer only the most serious cases to the IOP but that it was likely there were serious undiscovered cases to which the watchdog had not been alerted.

“We’ve got two investigations open,” Watson said. “Will that be the only case of misuse of AI? I don’t believe it will be. It’s a very tempting tool.”

FT research with forces has uncovered a small number of further cases not reported to the IOPC. In one of the most serious, a newly recruited officer in Sussex Police quit after being placed under misconduct investigation for using AI to help prepare a witness statement, and then denying it when challenged.

Watson said her organisation was encouraging forces to tell it about AI misuse cases so that it could advise whether the matter required an independent IOPC investigation or could be dealt with internally.

The IOPC has referred the case of the Derbyshire Detective Constable back to the force for it to conduct its own criminal investigation.

Watson stressed that it was not for her to decide what forces’ AI policies should be. But putting appropriate guidelines in place for AI was vital to ensuring the 43 overstretched police forces in England and Wales could use the systems when appropriate to speed up investigations and cut down on paperwork, she said.

“I think there’s a genuine moral imperative to embrace technology that clearly has potential benefits for productivity,” Watson said, adding that “sensible judgment and sensible policies” were also required.

PoliceAI said it was working with partners including the IOPC to support “consistent standards, training and oversight”.

It added: “We recognise that public confidence depends on AI being used responsibly, with appropriate safeguards in place and human accountability for decision-making.”