If a computer tells a court that a face matches a suspect, how should a jury interpret that claim? A confident-looking output is not the same as certainty, and misunderstanding the difference could have serious consequences.
British researchers are taking that problem into a realistic courtroom setting. In an announcement published on 9 November 2026, Northumbria University outlined mock criminal trials designed to explore how people understand and use AI-generated evidence.
The encouraging news is not that AI has been proved safe for the justice system. It is that researchers are examining the human decisions around it before assuming the technology deserves trust.
A Fictional Case, Real Legal Expertise
Researchers from Northumbria, Newcastle and Glasgow universities are staging the trials to examine responses from jurors, lawyers and judges. The scenario is fictional but realistic, with actors playing witnesses and the defendant.
The first trial is scheduled for Glasgow on 11-12 November, followed by a second in Newcastle upon Tyne on 16-17 March. The announcement gives those March dates without specifying the year. For this 10 November update, the Glasgow event is still ahead, and the announcement contains no main-trial results.
For the Scottish trial, 15 members of the public have been recruited to act as mock jurors and deliberate together. Real legal professionals will participate, with former senior Scottish judge Colin Sutherland, Lord Carloway, presiding. An expert in biometric systems and a serving police officer will also give evidence.
A pilot has already taken place in Northumbria’s moot courtroom with law students taking the roles of judge, counsel and jurors. That preparatory exercise should not be confused with completed results from the main research trials.
What Does an AI Output Actually Mean?
The research considers evidence including facial recognition and large language model outputs. The central question is not simply whether participants like AI, but how they interpret uncertainty and whether the technology changes their treatment of other evidence.
A probabilistic system can produce false positives and false negatives. Its output therefore needs context about what was measured, the conditions in which it was tested and what the result can reasonably support.
Researchers want to explore whether jurors over-rely on AI, dismiss it regardless of its quality or misunderstand its probabilistic nature. Lawyers may also need training and support to question and explain such material effectively. These are research questions, not conclusions that the mock jurors have already behaved in a particular way.
Looking Beyond the Moment Evidence Reaches Court
The project will examine preparation by legal professionals as well as what happens during the trials. That matters because an output may pass through several people or systems before a jury hears about it.
Errors or assumptions can accumulate along that chain. Observing how advocates present the material and question an expert could help identify where uncertainty becomes obscured or a limited claim is made to sound stronger.
The trials form part of PROBabLE Futures, short for Probabilistic AI Systems in Law Enforcement Futures. This is a four-year, £3.4 million Responsible AI UK-funded programme led by Northumbria, involving six universities. The £3.4 million supports the wider programme, not just these two mock trials.
Why This Is Worth Welcoming
The work could inform future guidance, professional training and rules for handling AI evidence. Whether it does so effectively will depend on the findings and how they are evaluated, rather than on the ambition of the announcement.
For anyone interested in responsible AI, this is a useful reminder that reliability is not only a technical score. People must also understand a system’s limits when making consequential decisions. Testing that understanding is a meaningful step towards better-informed justice.
Sources and Further Reading
- Northumbria University: Pathbreaking mock trials to test how AI evidence affects justice. Published 9 November 2026. Original university announcement verifying the event dates, participants, pilot, research questions and £3.4 million wider programme.