As artificial intelligence becomes more embedded in fraud investigations, a pressing question is emerging for counter-fraud professionals, investigators, and legal practitioners: when AI has played a role in developing a case, how does that need to be disclosed - and what are the courts likely to require?
A recent webinar brought together practitioners and legal experts to discuss this question directly. The answers were nuanced, but several clear principles emerged.
The Witness Must Understand the Evidence
Nick Jennings, Head of the Hertfordshire Shared Anti-Fraud Service, set out what he sees as the core principle. When evidence has been developed using AI - whether to identify patterns, generate leads, or assist with document review - the person presenting that evidence must be able to explain how it came about.
Nick described the role of a witness as being to present their evidence - evidence that they understand and can account for. "You can't just rely on information produced by a computer, whether it's AI or any other computer," he said. "You've got to be able to understand how you've got a piece of evidence, where it's come from, where the reliability is."
He added a practical note: when a question arises about how AI-derived evidence should be disclosed in a specific case, the right course is to refer it to a legal expert - not to assume that a screenshot of an AI output will be sufficient.
Witness Statements Are a Particular Concern
Akber Datoo, founder of D2 Legal Technologies and co-chair of the Technology and Law Committee at the Law Society of England and Wales, is currently involved in the Law Society's response to the Civil Justice Council's consultation on AI and court documents. He explained that the consultation covers pleadings, skeleton arguments, disclosure, witness statements, and expert reports.
For most of those document types, Akber said, the key question is who is taking professional responsibility and whether the material has been properly checked. But witness statements raise a distinct concern.
He explained that the Civil Justice Council's paper recognises that witness evidence needs to be in the witness's own words and drawn from their own knowledge. The concern with generative AI is that it may smooth away precisely the features of a witness account that carry important evidential weight - the hesitations, the qualifications, the expressions of uncertainty. AI could, he said, dilute hesitation, strengthen language in ways that do not reflect the witness's actual confidence, or make different witnesses sound artificially consistent. All of that goes directly to evidential integrity.
Akber said the position in criminal proceedings should be at least as cautious as on the civil side, and probably more so. The civil courts are dealing with rights, money, injunctions, and status. The criminal courts are dealing with conviction, acquittal, reputation, and liberty. "The criminal bar should be even less tolerant of AI generating, embellishing, diluting, or rephrasing the substance of a witness's account," he said.
Where AI Does Have a Legitimate Role
Neither Nick nor Akber suggested that AI has no place in criminal investigations - quite the opposite. Akber set out a range of ways in which AI can legitimately assist: transcription, formatting, indexing, triaging, building chronologies, reviewing disclosure, and pointing investigators towards relevant material within large datasets.
The critical line, he said, is between AI helping to find or organise evidence, and AI manufacturing, reshaping, or directing what that evidence is. And he was candid that the line is not always obvious. "It's a spectrum," he said. "We're not thinking about it properly if we think there's going to be some big sound when we cross over that line."
What Disclosure Actually Requires
An audience member asked a specific practical question: if AI has been used in the course of an investigation, is it simply a matter of exhibiting a screenshot of what was inputted and what the AI returned?
Akber's answer was that a screenshot might form part of the audit trail, but it does not come close to satisfying the requirement on its own. Where AI has materially influenced an investigation or the evidence being relied upon, he said, there is an obligation to preserve and be able to explain: the tool used, the data it was applied to, the prompt, the configuration, the output, the human review that followed, the limitations of the system, and the investigative step that resulted from it.
He drew a distinction between AI as a generator of leads and AI as part of the evidential basis of a case. If AI was simply pointing an investigator towards material - and the investigator then gathered and assessed that material themselves - the AI output may not itself be evidence. But the process may still need to be recorded and potentially disclosed if it affects the issues in the case.
"If AI has scrutinised the evidence in a way that influences what is relied on or what is excluded," Akber said, "then the defence, the court, and the process more generally needs enough transparency to be able to test that process."
A Consistent Principle
Rob Savage, VP for Public Sector at NUIX, brought the point back to a consistent theme across the whole discussion. "The question isn't really whether AI can do something," he said. "It's whether we can trust it to do it in a way that's fair, explainable, and defensible."
Public sector organisations, Rob noted, operate under a higher burden of accountability than most private organisations. The standard is not whether something works - it is whether it can be explained, defended, and evidenced.
Akber summarised the overarching principle clearly: if you are going to use an AI tool in a way that may be relied upon in court proceedings, you need to understand what the tool does, how it operates, and what the governed workflow around it looks like. "AI is not evidence," he said. "At best, it's a route to evidence and helping us with the process."
Head of the Hertfordshire Shared Anti-Fraud Service (SAFS)
Founder and CEO of D2 Legal Technologies and Professor of Law, Technology and AI at the University of Surrey
VP Sales - UK&I Public Sector, Nuix
This post is based on a webinar on "AI and fraud prevention in the public sector", featuring speakers from the Public Sector Fraud Authority, the Hertfordshire Shared Anti-Fraud Service, NUIX, and D2 Legal Technologies. Listen to the whole thing for free here >> https://register.govnet.co.uk/webinar-ai-vs-fraud
Jessica Kimbell, GovNet
