Judge Dismisses Claim After AI Hallucinates Witness Statement

In a significant intervention regarding the intersection of technology and the justice system, a senior employment judge in Northern Ireland has issued a stark warning about the use of artificial intelligence in legal proceedings. Employment Judge Sheils described a claimant’s reliance on generative AI to draft a witness statement as a “disturbing development,” highlighting the severe risks posed by automated text generation in a tribunal setting.

The comments were made as part of a written ruling concerning a disability discrimination case lodged against the Northern Ireland Fire and Rescue Service. The case has drawn attention not merely for its subject matter, but for the procedural anomalies introduced by the claimant, who was representing himself as a litigant in person. The tribunal’s findings serve as a cautionary tale for individuals attempting to navigate complex legal landscapes with the assistance of unverified digital tools.

The Dangers of Artificial Hallucinations

The core of the tribunal’s concern lay in the accuracy and integrity of the evidence presented. During the proceedings, it emerged that the claimant had utilised an AI tool to assist in preparing his witness statement. While the claimant insisted that the technology was used solely to analyse legal points, the outcome was a document that the judge found to be fundamentally flawed.

Generative AI models are known to suffer from “hallucinations”—confident assertions of fact or law that are entirely fabricated. In this instance, the technology significantly misled the claimant regarding a pivotal piece of case law. The AI provided an erroneous interpretation of Dornan v Chief Constable of the Police Service of Northern Ireland (2022), suggesting the outcome was favourable to the claimant’s position when, in reality, it presented substantial hurdles.

Judge Sheils noted: “The claimant stated that he had used AI only for the analysis of the legal points but, as it transpired, AI unhelpfully got a significantly important case law wrong, which could have been seriously misleading the claimant as to his chances of succeeding in his claim.”

This error was particularly egregious given that a previous employment judge had specifically directed the claimant to that very case during a preliminary case management meeting, intending to warn him of the legal difficulties he faced. Instead of heeding the judicial guidance, the claimant relied on the algorithm’s incorrect summary, claiming he had been unable to locate the case himself.

Blurring the Lines Between Fact and Argument

Beyond the misinterpretation of legal precedent, the use of AI contributed to a procedural mess regarding the witness statement itself. By strict tribunal rules, a witness statement must stand as the claimant’s sworn testimony, containing factual evidence relevant to the issues at hand. It is not the place for legal arguments or subjective submissions.

However, Judge Sheils observed that the document produced was “scant” on factual evidence and instead contained a “confusing meld of fact, opinion and legal argument.” This failure to separate evidence from submission created significant difficulties for the tribunal and the respondent. It also forced the claimant to add oral evidence during cross-examination that had been omitted from his written statement, further complicating the hearing.

“The tribunal found this to be a very disturbing development as, by rule, witness statements are adopted as the claimant’s own sworn testimony,” the judge remarked, underscoring the sanctity of the witness box which sits at odds with the synthetic output of a chatbot.

Outcome and Implications

Ultimately, the reliance on technology could not save the claim. The tribunal found that the case had been brought out of time. Despite the claimant’s arguments, Judge Sheils concluded that the claimant had failed to establish that it would be “just and equitable” to extend the statutory time limits. Consequently, the claim was dismissed for want of jurisdiction.

This ruling reinforces a growing global judicial consensus: while technology can assist in legal research, it cannot replace the nuanced understanding of a qualified professional, nor can it substitute for the personal, sworn recollection of a witness. Litigants in person are now on notice that the courts will scrutinise AI-generated content with rigour, and that “the computer said so” is no defence for misleading a tribunal.