A 300-page ChatGPT-generated skeleton argument failed to comply with EAT requirements.
The Employment Appeal Tribunal (EAT) has warned litigants that they remain personally responsible for checking AI-generated legal documents before submitting them, after an unrepresented appellant filed a 300-page skeleton argument created using ChatGPT. The judgment in Hancox v Sutherland & Others [2026] EAT 139 was handed down on 17 September 2026.
The case concerned an appeal by Terence Hancox after an Employment Tribunal struck out claims against four individual respondents. The tribunal had found that the proceedings had been conducted in a “scandalous, unreasonable or vexatious” manner and that a fair trial was no longer possible.
At the EAT preliminary hearing, Hancox, assisted by an ELAAS representative, pursued one ground of appeal: whether he had received fair notice that the strike-out application would be determined and a fair opportunity to prepare and participate.
The EAT held that this ground was reasonably arguable and directed that it proceed to a full hearing. However, Hancox subsequently sought to introduce another ground concerning his rights under Articles 6 and 10 of the European Convention on Human Rights. The judge dismissed that application, finding that he had clearly confirmed at the hearing that only the first ground was being pursued.
A significant part of the judgment addresses the use of generative AI in litigation. The EAT recognised that AI can assist people who do not have access to professional legal advice, but warned that its output can be inaccurate, excessive, unfocused or procedurally non-compliant. The court noted particular risks including fabricated case citations, incorrect legal propositions and material that does not reflect the evidence or legal position.
The judge said the 300-page skeleton argument submitted by Hancox did not comply with the EAT’s Practice Direction and had served to obscure rather than clarify the issues. The judgment emphasised that skeleton arguments should generally be between five and 15 pages and should identify the relevant legal points, authorities and steps in the argument clearly.
The EAT stressed that these requirements apply to represented and unrepresented litigants alike. Anyone using generative AI must ensure that documents comply with procedural rules, check factual, evidential and legal material for accuracy, and remove irrelevant or repetitive material. Simply submitting AI-generated material without checking it is not acceptable.
The court also dismissed Hancox’s disclosure application as totally without merit and rejected applications to review earlier EAT orders. It ordered that future applications be made using the prescribed Practice Direction form and warned that continued excessive correspondence could lead to restrictions on his ability to communicate with the EAT by email.
The appeal itself will proceed on the single ground the EAT found reasonably arguable, with further case-management directions made for the full hearing.