Tuesday 26 May 2026 4:15 am

| Updated:

Monday 25 May 2026 1:45 pm

AI-hallucinated evidence in court has become a structural issue that needs a procedural response, writes Nicholas BlomfieldThe courts are about to be stress-tested in a way they have not evolved to cope with. Not by volume or funding, but by something more fundamental; whether the evidence presented to them is trustworthy. AI-generated material has been in circulation for a few years and there is now a string of notable cases. This includes R (on the application of Ayinde) v London Borough of Haringey, where AI-generated evidence was submitted to the court which turned out to be fabricated case law, otherwise referred to as “hallucinations”. The High Court imposed a wasted costs sanction and referred both the barrister and solicitor to their respective regulators, while Mr Justice Ritchie described the conduct as “grossly unprofessional”. Dame Victoria Sharp P went further by reinforcing that unverified reliance on AI constitutes negligent conduct, and warned that the integrity of the justice system depends on genuine, informed human oversight rather than superficial review.AI has become pervasive in legal industryAI material is becoming pervasive across legal workflows, in witness statements, in disclosure reviews, contracts, policies and even in expert witness reports. Harvey AI and Legora are leading the charge with large language models (LLMs) and we are about to see a tsunami of legal materials hit the market as a result. There are multiple issues that will come with this, but one intrinsic issue is whether a court should assess, and rely on, evidence that may be partly or wholly generated by a system capable of producing authoritative but, in some cases, false outputs. In Ayinde, fabricated case law and materials were submitted to the High Court with dire consequences for the legal advisors. The civil justice system in England and Wales relies on assumptions that are quickly becoming unpredictable and, worse, unreliable. The current legal system is built on human accountability. A witness signs a statement of truth, a solicitor drafts and certifies documents, or an expert witness produces a court guiding report. Each of these assumes a traceable human contributor behind the content. AI has materially disturbed the equilibrium, introducing models that infer facts, fill gaps, re-write statements or reshape language based on material and assumptions that may be inherently wrong. The outputs may appear coherent, commercial and accurate but in fact may be flawed in ways that are difficult to detect without investigation. Hallucinations will become the norm, not exceptionFabricated cases, as in Ayinde, are an issue, but the real risk is normalisation as AI becomes part of routine drafting and document handling. Contamination of evidence will not be exceptional, it could well become routine. This creates several procedural problems including authenticity, attribution and weighting. In terms of authenticity, courts are accustomed to distinguishing genuine documents from forgeries, but AI introduces a further category of documentation, a hybrid synthetic that introduces a convincing creation. Our current justice system and evidential frameworks are not equipped to handle these creations or to know definitively if legal weight and mechanisms should be placed on them.Then, when it comes to attribution, amid the explosion of AI, responsibility is rapidly becoming difficult to ascertain. Is it the platform that produces, the solicitor who is the ‘conduit’ or the customer who relies? Existing duties under the Civil Procedure Rules are framed around human behaviour, they are not designed to apply to probabilistic autonomous systems. Even where AI use is acknowledged, it provides very little to the court. A statement that AI was used “for drafting assistance only” is meaningless unless it addresses the critical issue as to the extent to which the system influenced substance and direction. The procedural challenges are becoming more evident, particularly in how evidence was gathered and created. Inevitably this will lead to increased costs, further delays and additional pressure on an already stretched system. Case-by-case change won’t cut itThis is a structural issue, and it requires a procedural response. It is not a problem that can be resolved through incremental case-by-case guidance. Some potential risk mitigation could involve the use and extent of AI should be disclosed at the start of any client engagement and certification obligations on legal representatives should be mandatory i.e. clear confirmation that reasonable steps have been taken to check and verify the accuracy and contents of any AI-produced materials. Courts should be prepared to draw adverse inferences where AI use is concealed or inadequately explained. When used properly, AI has the potential to improve efficiency, particularly in disclosure, drafting and analysis; however, evidence is the foundation on which decisions are made. If that is unstable or unreliable, everything built on it is at risk, most notably, precedent. The judicial system has modernised in many different ways and for varying reasons. For example, online hearings became the norm during Covid-19, albeit they are considered less favourable now and the system has largely reverted to the original in-person hearings. The question here is not about the method but the material, and whether the courts can truly be asked to trust it. If the court cannot distinguish between evidence grounded in fact and evidence generated by machine inference, the process of fact finding is compromised at its source. At that juncture, it is no longer a procedural issue but whether, as Dias J stated when referring Ayinde to the High Court, the courts can actually rely on the integrity of the people and submissions presented before it. Moreover, as Dame Victoria Sharp P described in her judgment, referring to Dias J’s earlier comments, the question is whether the system with the influence of AI can deliver actual justice and the “public confidence in the administration of justice” can still be maintained.Nicholas Blomfield is a disputes lawyer working across technology and digital assets