Hancox v Sutherland: EAT sets out rules for AI use in tribunal litigation

The Employment Appeal Tribunal has set out detailed guidance on litigants' responsibilities when using generative AI to prepare court documents, in a judgement that also allowed part of an appeal against a strike out order to proceed to a full hearing.

The underlying claim was brought by Terence Hancox against the National Farmers' Union and, in these proceedings, four of its individual employees, following his dismissal in February 2020. Hancox alleged he had made protected disclosures about health and safety matters and was automatically unfairly dismissed as a result. In April 2021, an Employment Tribunal struck out his claim against the four individuals, finding that a sustained LinkedIn campaign criticising the respondents, including the sharing of documents disclosed during litigation, amounted to scandalous, unreasonable and vexatious conduct that had made a fair trial impossible.

Hancox appealed, and a preliminary hearing was listed before Marcus Pilgerstorfer KC, sitting as a Deputy Judge of the High Court, to determine whether any grounds disclosed a reasonable basis for the appeal to proceed. Ahead of that hearing, Hancox, acting as a litigant in person, filed a purported skeleton argument running to 300 pages and roughly 132,000 words, which he confirmed had been generated using ChatGPT. The judgement described the document as entirely unacceptable, noting it bore no resemblance to the concise submissions required by the tribunal's Practice Direction and that Hancox had, in effect, invited the respondents to identify any inaccuracies within an unreasonably short timeframe rather than checking the content himself.

Drawing on recent authority including R (Ayinde) v London Borough of Haringey, the judgement set out clear expectations for all litigants using AI tools, whether represented or not: documents must comply with procedural rules, be checked as thoroughly as reasonably possible for factual and legal accuracy, and focus on the strongest points without undue repetition. The judge emphasised that responsibility for checking AI-generated material could not be shifted onto an opposing party, and that non-compliant documents risk rejection or other sanctions.

On the substance of the appeal, with the assistance of pro bono ELAAS counsel at the hearing, Hancox pursued a single ground: that the original tribunal had struck out his claim without giving him fair notice that the application would be determined, or a fair opportunity to prepare a response. The judgement found this reasonably arguable, noting that correspondence from the tribunal in the days before the hearing could reasonably have been read by a litigant in person as meaning the strike out application would merely be discussed rather than decided, and that documents were provided close to the hearing date.

After the hearing, Hancox sought to revive a second ground concerning whether his rights under Articles 6 and 10 of the European Convention on Human Rights had been properly weighed. The judgement refused permission to resile from the clear position taken at the hearing, finding no good reason for the change of position and that the point was in any event not reasonably arguable, since the tribunal had plainly considered the competing fair trial rights of both sides.

The judgement also addressed what it described as an unusually large volume of correspondence and repetitive applications from Hancox, some acknowledged by him to be linked to his mental health. While declining to restrict his communications with the tribunal at this stage, the judge warned that continued excessive correspondence could result in an order limiting him to communicating by post. A disclosure application was dismissed and certified as totally without merit.

Case management directions were given for the case to proceed to a full hearing on the single surviving ground.