Summary

• The Home Office has accepted key recommendations from Jonathan Fisher KC’s Independent Review of Disclosure and Fraud Offences.
• The reforms would allow AI to help police review and summarise evidence in criminal cases.
• The move links public sector AI adoption to a concrete operational bottleneck in policing, prosecution, and court readiness.

The Home Office has accepted key recommendations from Jonathan Fisher KC’s Independent Review of Disclosure and Fraud Offences, including changes that would allow AI to help police review and summarise evidence in criminal investigations.

The reforms are aimed at disclosure rules introduced in 1996, when criminal case files were far more likely to be paper based and far less likely to contain mobile phone downloads, cloud records, chat histories, images, videos, and large quantities of financial or communications data. Ministers say some investigations now involve the equivalent of more than 500,000 e-books of material, while an average fraud case can contain more than four million documents.

Disclosure is one of the least glamorous parts of the justice system, but it has become one of its most important digital bottlenecks. Police and prosecutors must identify material that may be relevant to the case, including evidence that could assist the defence or undermine the prosecution. As investigations have become more data heavy, the process has created heavy administrative pressure on officers, lawyers, and courts.

AI is being proposed as a way to review, organise, and summarise large volumes of digital material so that officers spend less time handling paperwork and more time on investigative work. The government also plans to use Home Office funding for PoliceAI to pilot tools that can automatically generate summaries of digital material, with a view to scaling across forces in 2027.

Criminal disclosure is not a conventional document management problem. Evidence must be handled lawfully, securely, and transparently, while defence rights, privacy, proportionality, and auditability are protected. Any AI system used in the process will need to show how material was searched, prioritised, summarised, and escalated, particularly where decisions could affect whether a trial is fair.

The disclosure reforms are a useful test of public sector AI adoption because the operational pain is clear, the data volumes are measurable, and the consequences of poor design are serious. A tool that misses relevant material, produces unreliable summaries, or creates opaque case handling could damage trust rather than improve throughput.

Police forces will also need consistent rules on procurement, model evaluation, security, and staff training. Evidence data can include personal information, communications records, sensitive images, financial documents, and material involving victims, witnesses, and suspects. Systems used to process that information must be protected against unauthorised access and designed to avoid uncontrolled reuse of sensitive data.

The reforms sit alongside a wider government push to use AI in policing and justice. Recent initiatives have included the launch of PoliceAI and work on AI tools for courts, while ministers have also consulted on the use of biometrics, facial recognition, and related technologies in law enforcement. Taken together, those moves show a justice system trying to modernise around data volume while raising the need for clearer governance.

The largest implementation risk is that AI becomes a layer added on top of fragmented systems rather than a genuine redesign of evidence handling. Police forces, prosecutors, defence lawyers, courts, and digital forensics teams all touch parts of the disclosure process, and each organisation has different systems, pressures, and incentives. Unless reform addresses workflow, data standards, and accountability, software alone will shift the backlog rather than remove it.

The government’s decision to legislate for AI assisted disclosure still leaves open the harder operational questions: which tools will be used, how they will be tested, how errors will be challenged, and how human oversight will work in practice. The justice system needs help with digital evidence, but it also needs to prove that automation strengthens due process rather than making case preparation faster at the expense of confidence.