The rapid adoption of AI across the justice system has hit a speed bump after several UK police forces were instructed to stop using AI tools to prepare court statements and undertake other criminal justice tasks.

The intervention, led by Alex Murray, head of the newly established Police.AI centre, reflects growing concerns that inaccurate or fabricated AI outputs could undermine criminal investigations, contaminate evidence, and ultimately jeopardise legal proceedings.

While AI continues to promise major efficiency gains for police forces and legal professionals, the warning serves as a reminder that when criminal liability and the integrity of court proceedings are at stake, accuracy cannot be sacrificed for convenience.

A warning from the front line of criminal justice

According to reports, several police forces in England and Wales had begun using commercially available generative AI tools to assist with converting officer interviews into court statements and preparing disclosure-related documents.

However, Murray confirmed that Police.AI had intervened in some cases, instructing forces to halt deployment until proper assessments and safeguards had been completed.

His reasoning was that any technology used within the criminal justice system must meet a standard of reliability that is effectively “beyond reasonable doubt.”

That threshold is higher than the level of accuracy typically expected in commercial or administrative settings. A minor error in a business report may be inconvenient. An error in a witness statement, disclosure schedule, or document could affect the outcome of a criminal trial.

The concern is not theoretical. Last year, West Midlands Police reportedly relied on AI-generated material produced by Microsoft’s Copilot that fabricated details of a historical football match involving Maccabi Tel Aviv. The false information found its way into a police dossier supporting a proposed ban on supporters attending a match.

The incident became a high-profile example of what are commonly known as AI “hallucinations,” situations where AI systems generate convincing but entirely false information.

Why this matters for law firms

For criminal defence solicitors, barristers, prosecutors, and law firms, the implications are profound. Traditionally, witness statements have been viewed as evidence reflecting a witness’s own recollection and account of events. However, if AI systems are being used to draft or summarise those statements, new questions arise regarding authorship, reliability, and integrity.

Lawyers may increasingly need to ask:

• Was AI used in the preparation of this witness statement?
• What AI system was used?
• How much of the final statement was generated or altered by AI?
• What safeguards were in place to verify accuracy?
• Could the AI have introduced errors, omissions, or misleading language?

Defence teams may also seek disclosure relating to AI-assisted drafting processes where there are concerns that the technology influenced the content of evidence presented before the court.

In criminal proceedings, even small factual inaccuracies can have significant consequences. If AI-generated text subtly changes the wording of a witness account, omits important context, or introduces facts not originally stated, the reliability of that evidence may be challenged.

The risk of AI-generated witness statements

The issue extends beyond policing. As generative AI tools become more accessible, legal practitioners should be alert to the possibility that witnesses, defendants, or even professional advisers may use AI to help draft statements before they are submitted to the court.

An AI-generated witness statement may appear polished but it may not accurately reflect the witness’s own recollection. Worse, it may contain details suggested by the AI that were never actually remembered or experienced by the witness.

This creates potential challenges around credibility. Courts have scrutinised witness coaching and the preparation of evidence. AI introduces a new dimension to that discussion. A statement that has been substantially shaped by generative AI may raise questions about whether it genuinely represents the witness’s own account.

Criminal practitioners should therefore consider discussing AI usage directly with clients and witnesses during evidence preparation and review processes.

Disclosure and due diligence matter

Particular concerns have also been raised regarding disclosure schedules or the records of material gathered during investigations that must be disclosed to the defence.

Disclosure failures have been responsible for many miscarriages of justice and collapsed prosecutions over recent decades. Introducing AI into the process creates additional risks if relevant material is incorrectly categorised, summarised inaccurately, or overlooked altogether.

For law firms, this reinforces the importance of understanding how AI systems are being used by police, prosecutors, expert witnesses, and clients.

The focus is likely to shift from simply asking whether AI was used to understanding how it was used, what controls existed, and whether outputs were independently verified.

AI is not the problem, governance is

The Police.AI intervention does not mean there is firm opposition to AI. Police leaders highlight promising use cases, including reviewing CCTV footage, identifying suspects more quickly, and assisting with the analysis of large volumes of digital evidence. AI may also help reduce officers’ exposure to harmful material in investigations involving child sexual abuse imagery.

These applications demonstrate the real benefits that AI can deliver when deployed appropriately and with adequate safeguards.

Police.AI is not saying don’t use AI. But it is emphasising that AI cannot be used without proper governance, testing, oversight, and accountability. That distinction is crucial for law firms, legal departments, and criminal justice professionals.

A new era of AI in the courts

The police decision may represent one of the first major examples of a public authority actively slowing AI adoption because of evidential and legal concerns rather than accelerating deployment for efficiency gains.

As AI becomes increasingly embedded within investigations, witness preparation, document review, and evidence management, lawyers will need to develop a deeper understanding of how these systems operate and where their risks lie. Questions about AI-generated evidence, AI-assisted witness statements, and AI-supported disclosure processes are likely to become increasingly common in criminal proceedings.

In criminal law, where the standard remains beyond reasonable doubt, AI-generated content can never be accepted at face value. Every output must be treated as a starting point for verification, not as evidence in its own right.