Ministry of Justice adds AI-enabled crime to its AI action plan, a year after launch

The Ministry of Justice says it is adding a fourth strategic priority to its AI Action Plan for Justice, aimed at "emerging AI risks, including AI-enabled criminality". The change is set out in a one-year update published on Thursday 24 September, which reviews progress since the plan was launched in the summer of 2025.

In a foreword, Jake Richards MP, the department's lead minister for AI, writes that "as AI capabilities continue to advance, so too do the risks they present, and we must be ready to respond."

The new priority

The update says the risks include "the misuse of AI to support criminal activity, the growing prevalence of synthetic and manipulated content, and increased demand on the justice system as AI tools change how citizens engage with justice services". It says these affect organisations across the justice system "in different but interconnected ways, requiring a coordinated, cross-system response".

Two actions sit under it:

• 4.1: to identify, assess and escalate emerging AI risks, "building a more systematic horizon-scanning capability", and to improve understanding of AI-enabled criminality and "how it may change patterns of offending and affect demand for justice services".
• 4.2: to develop "tools, guidance, safeguards and broader strategies as appropriate" with partners; the document says this "may include work to address the increased risk of synthetic or manipulated content entering the justice system".

The update does not give a budget, timetable or named deliverables for the new priority.

What the department says it has done

The update says the department's flagship tool, Justice Transcribe, which uses speech recognition to transcribe, summarise and structure records of probation supervision sessions, has been scaled to over 12,000 probation officers. It says "over 1.5 million meetings were summarised between 7 October 2025 and 14 September 2026", and that the tool has also been rolled out nationally to all judges in the First-tier Tribunal (Immigration and Asylum Chamber). The department says it is exploring expanding it to other areas, including courts.

It also lists non-generative tools it describes as live across probation, prisons and courts:

• a Violence in Prisons Estimator, which "uses predictive analytics to assess the risk of involvement in violent incidents in custody";
• Acquisitive Crime Mapping, which "matches GPS location data of people on probation with police crime data";
• Online Probation Check-ins, which "uses facial recognition to support identity verification of people on probation".

Beyond the original plan, the document says a Digital Rapid Response Unit built AI-enabled tools to help prevent prisoners being released in error, including data-linking tools aimed at offenders who use multiple aliases. It says the department is also developing AI-enabled tools to identify case-readiness issues and support listing decisions in the Crown Court backlog, and will work with industry to develop "robust benchmarks and safe evaluation standards for AI in legal settings".

On suppliers, the update names partnerships with Microsoft and OpenAI, and says the Ministry of Justice was "the first government department to roll out an enterprise version of ChatGPT"; that claim is the department's own. Staff have access to an enterprise version of Copilot Chat, it says, alongside ongoing rollouts of Copilot for Microsoft 365 and ChatGPT Enterprise. For year two it says it will "explore options for enterprise-wide agreements with further vendors to support a more diverse, multi-vendor AI portfolio".

Year-two plans and what is not known

Other year-two priorities include scaling "agent‑based tools for discrete, repetitive tasks", expanding access to approved AI coding tools to modernise legacy systems, and piloting AI interventions on the Crown Court backlog. The department also says it will "develop a clearer view of how AI will affect future workforce needs", and that future work will follow consultation requirements, "including with trade unions where relevant".

The update says all AI tools must be tested for "accuracy, bias, fairness, security and reliability before widescale deployment", but it does not publish those evaluation results for the tools it names. It points readers to a separate interactive "AI Action Plan Explorer" for detail on individual actions, which we have not reviewed.

On our reading, the update is mostly a record of adoption, and the new fourth priority is the part that turns the department's attention from using AI to the AI-related risks facing the justice system.