The claimant, a litigant in person, relied on a skeleton argument generated using ChatGPT. According to the judgment, the document ran to approximately 300 pages and around 132,000 words. The EAT considered the resulting procedural difficulties in some detail and reiterated the responsibilities that arise when AI is used to prepare material for litigation.

While the decision arose in the context of employment proceedings, the issues considered by the Tribunal will be familiar to many organisations already dealing with AI-assisted complaints, grievances and disclosures.

While the decision arose in the context of employment proceedings, the issues considered by the Tribunal will be familiar to many organisations already dealing with AI-assisted complaints, grievances and disclosures.

Interim Relief Appeal

The proceedings arose from an appeal against an Employment Tribunal decision. Among the remedies sought by the claimant was interim relief, a jurisdiction intended to provide urgent protection in a limited category of cases.
Against that background, the EAT was required to consider extensive written submissions which far exceeded the scale ordinarily expected in appellate proceedings.

The Tribunal observed that the skeleton argument did not comply with the relevant procedural requirements and did not perform the function expected of a skeleton argument, namely identifying and explaining the issues requiring determination.
The scale of the document is a significant feature of the judgment.

What Concerned the Tribunal?

The judgment is not a warning against the use of AI as such.

The EAT acknowledged that AI tools can assist litigants and can play a role in access to justice. However, the Tribunal also referred to well-known risks associated with generative AI, including inaccurate legal propositions, fabricated authorities and the production of material that requires careful verification.

Of equal interest is the Tribunal's concern about volume.

A recurring theme in the judgment is that the sheer quantity of material generated did not advance the litigant’s case or assist the efficient determination of the appeal. The Tribunal repeatedly returned to the need for relevant, focused and procedurally compliant submissions.

This issue will resonate beyond litigation. One consequence of generative AI is that it has become much easier to produce bulky, legalistic-sounding written material. The challenge for decision-makers is no longer obtaining information, but separating the meaningful and relevant from the irrelevant. Any relevant points made within the prompting are, at best, obscured, or, at worst, entirely lost.

Responsibility Remains With the User

A clear message from the judgment is that the use of AI does not alter responsibility for the finished product. This is consistent with the recent SRA Warning Notice on AI and Judicial AI Guidance.

The Tribunal rejected any suggestion that shortcomings in AI-generated material could be attributed to the technology itself or left for opponents to identify and correct. The obligation to ensure that a document is accurate, relevant and compliant remains with the person who submits it.

These principles are familiar; what is new is the AI-enhanced landscape to which they are being applied.
As AI tools become increasingly omnipresent, courts and tribunals are likely to encounter more submissions prepared wholly or partly using generative AI. Hancox suggests that traditional expectations regarding accuracy, relevance and procedural compliance will continue to apply regardless of how a document is produced.

Practical Ways to Address the Challenge of AI-Generated Volume

Although the case concerned litigation, many will have now experienced similar issues, albeit not to the 300-page extreme.
HR and legal teams are increasingly receiving grievances, whistleblowing disclosures and complaints that appear to have been drafted or expanded with the assistance of AI tools. Documents that might previously have been a few pages long can become significantly longer, often combining factual allegations, legal argument, commentary and repetition.
A five-page letter will attract a ten-page response, which will attract a fifteen-page reply.

The overall result is cost, delay and increasing entrenchment, without any real narrowing of the issues.
One practical lesson is that organisations should not feel compelled to respond to AI-generated volume with more AI-generated volume. In some cases, an early conversation may do more to identify the issues in dispute than a succession of increasingly lengthy letters. The objective should be to understand and investigate the substance of the complaint, not simply to win a battle of word counts.

The judgment provides some useful lessons. When dealing with a lengthy AI-assisted complaint or disclosure, organisations may wish to:

• Identify and clarify the allegations requiring investigation at an early stage;
• separate factual assertions from commentary and legal argument;
• group overlapping issues together where appropriate;
• maintain a clear record of how allegations have been categorised and considered; and
• focus investigative resources on matters that are genuinely material to the outcome.
  
  

The existence of a lengthy document does not reduce the need for a thorough investigation. Equally, volume should not be mistaken for relevance.

Conclusion

The unusual facts of Hancox have given rise to one of the more detailed judicial discussions of AI use in employment litigation to date.

The judgment does not suggest that AI should be avoided. Nor does it create a separate set of rules for AI-generated documents. Instead, it applies familiar procedural principles to a new technology.

For litigants, lawyers and employers alike, the lesson is a practical one. AI can generate words almost without limit. The challenge for lawyers, HR professionals and decision-makers is increasingly not producing material, but identifying the issues that genuinely matter. Sometimes the response to AI material may begin not with another written response, but with a telephone call.