AI-generated court documents: lessons from a recent EAT decision

A litigant in person had filed a 300-page skeleton argument created using ChatGPT with the Employment Appeal Tribunal (EAT) as part of an appeal of a strike-out order made by an employment tribunal. He accepted that the EAT could not realistically read it before the hearing and invited the respondents to identify any factual errors, inaccurate quotations, incorrect dates, faulty bundle references or misrepresentations.

The EAT found the document entirely unacceptable. It did not comply with the EAT Practice Direction 2024 (PD), served to ‘obscure rather than to illuminate’ the issues and was not within the PD's recommended five to 15 pages. The claimant had not checked it and accepted that it lacked credibility.

The EAT has observed a significant increase in AI usage before courts and tribunals and stated ‘As is well-known, the technology presents both risks and opportunities. There is potential benefit to be gained from its effective, responsible and ethical use, including by those who do not otherwise have access to professional legal assistance. Greater benefit is likely to arise as the technology improves and there is little doubt that "[a]rtificial intelligence is likely to have a continuing and important role in the conduct of litigation in the future. There are also risks, including (and I do not attempt an exhaustive list) that the output is unfocused, excessive or not of direct relevance; does not comply with procedural requirements; is the product of inadequate/inaccurate input material or poor prompting; is misleading or does not reflect the true factual, evidential or legal position in a case; raises new points on appeal that were not raised at first instance; is affected by in-built biases; or contains hallucinations (for example made-up case citations).’

The EAT made the point that compliance with the EAT's procedural requirements is mandatory, not optional and that in the present case the EAT was left with an unmanageable document which impeded preparation for the hearing. The skeleton argument was not relied upon at the hearing.

Although the litigant in person was given permission to appeal, the EAT warned that non-compliant AI-generated documents may be rejected and that other sanctions could follow where appropriate.

Learning points

The EAT used the case as an opportunity to address the growing use of generative AI in tribunal litigation and to clarify the responsibilities of those who rely on such tools. All litigants (in person and professional) who use AI must ensure, at a minimum, that all documents submitted:

• Comply with applicable procedural rules. In the present case, the 5-15 page recommended rule and any regulatory guidance. Use of AI does not excuse procedural non-compliance.

• Have been checked as thoroughly as reasonably possible for accuracy, ensuring that factual, evidential and legal points (including references to authorities) are correct and that the court or tribunal is not misled. Remember using AI is not a defence to inaccuracy.

• Contain only relevant points, focused on central or best arguments, presented in an easily comprehensible manner without undue repetition.

The EAT also made clear that it is not acceptable for a party to submit AI-generated material and then seek to place responsibility on an opponent to identify inaccuracies, errors or hallucinations. Responsibility for the content of court documents remains with the person submitting them.

The case is not a rejection of the use of generative AI in legal proceedings. Rather, it is a reminder to all court users that AI is an assistive tool, not a substitute for professional judgment. Organisations and legal teams adopting AI-assisted workflows should ensure that robust review and verification processes remain in place, particularly where documents are intended for submission to courts, tribunals or regulators.

Find out more about our Commercial Disputes expertise or contact us today to discuss how we can support you.