AI slop in a tribunal? What this means for employers using ChatGPT

Written by Sammy Sheikh from OOOM Global Limted

You've probably seen AI slop.

Scroll through social media for long enough and you will find it.

AI generated images that don't quite make sense. Articles that use hundreds of words to say very little. LinkedIn posts that sound polished but somehow feel empty. Videos, comments and content being generated at a speed that makes you wonder whether anyone actually stopped to check them before pressing publish.

"AI slop" has become an informal term for low quality AI generated content, particularly content produced quickly, at scale and with limited human scrutiny.

Usually, the consequence is that you scroll past it. But what happens when that same approach finds its way into something more serious? Like an Employment Appeal Tribunal (EAT).

That is where Hancox v Sutherland and others [2026] EAT 139 becomes interesting.

300 pages of ChatGPT

Mr Terence Hancox was a litigant in person, meaning he was representing himself.

As part of his appeal, he submitted a 300 page skeleton argument created using ChatGPT.

The EAT's usual guidance is that a skeleton argument should generally be between five and 15 pages. If it exceeds 20 pages, a judge may require a shorter version or reduce the time available for oral submissions.

Hancox's document contained almost 132,000 words. But size was not the only problem.

The EAT said the document did not comply with its Practice Direction and "served to obscure rather than to illuminate" when the Tribunal was preparing for the hearing.

The judge later described it as "entirely unacceptable".

At the hearing, Hancox explained that he had used ChatGPT because he needed to prepare the document quickly. He acknowledged that much of it lacked credibility.

The EAT found that he had not carried out the necessary checks to ensure that the document complied with procedural requirements or that its contents were accurate.

And there was another problem. Hancox had effectively attempted to put the responsibility on the respondents to identify inaccuracies in his document.

The EAT was clear: that was not acceptable. Responsibility remained with the person who submitted it.

Artificial intelligence shouldn't make you unintelligent

That might sound obvious, but it is probably one of the most important lessons businesses can take from this case.

ChatGPT can produce something that looks remarkably professional in seconds. It can quote legislation, explain legal concepts, structure an argument, and draft a letter that sounds as though someone spent hours preparing it.

And that creates a new problem. Something can look credible without necessarily being correct.

The EAT recognised that AI has genuine benefits. The judgment says generative AI has the potential to help people who might not otherwise have access to professional legal assistance and acknowledges that AI is likely to have an important continuing role in litigation.

So this wasn't an anti AI judgment, but a warning about uncritical AI use.

The EAT identified risks including AI output being unfocused, excessive or irrelevant; failing to comply with procedural requirements; being based on inaccurate information or poor prompting; misstating the factual or legal position; containing bias; and even producing hallucinations such as made up case citations.

In other words, AI can be extremely useful. But artificial intelligence shouldn't make you unintelligent.

The EAT has now made its expectations clear. The judgment goes further than simply criticising what happened in Hancox.

It provides guidance for people using generative AI to create documents submitted to the EAT. At a minimum, the EAT said users should ensure that documents:

• Comply with the applicable procedural rules.
• Are checked as thoroughly as the person is reasonably able for accuracy, including factual, evidential and legal points and references to authorities.
• Contain only relevant points, concentrating on the central or strongest arguments, presented clearly and without unnecessary repetition.

The judgment then makes another important point.

Simply submitting the product of generative AI or expecting the opposing party to check it for you is not acceptable.

Where adequate checks have not been undertaken, the EAT warned that judicial enquiries and potential sanctions may follow. That is quite a significant message.

What does any of this have to do with employers?

Potentially, quite a lot.

The specific guidance in Hancox concerns documents submitted to the Employment Appeal Tribunal. It does not create a new rule saying businesses cannot use ChatGPT for HR.

But the underlying problem is highly relevant to the workplace.

Managers and HR teams can now ask AI to help them draft:

• Grievance responses.
• Disciplinary letters.
• Investigation questions.
• Policies.
• Meeting notes.
• Redundancy communications.
• Performance management documents.
• Even employment law advice.

This can save enormous amounts of time.

The danger begins when "help me draft this" becomes "tell me what to do."

"ChatGPT says we can dismiss them"

Imagine a manager types: "My employee has been off sick six times this year. Can I dismiss them?"

Within seconds, ChatGPT produces a polished answer. Perhaps it explains absence management, gives the manager a process to follow and even refers to employment law.

It looks convincing. But what does the AI actually know?

• Does it know why the employee has been absent?
• Could the employee be disabled for the purposes of the Equality Act 2010?
• Has occupational health advice been obtained?
• Have reasonable adjustments been considered?
• What does the company's own absence policy say?
• How have similar employees been treated?
• What does the employment contract say?
• Has the manager even provided an accurate version of events?

Change one important fact and the appropriate course of action could change with it. AI cannot consider information it has never been given.

That is why a confident answer should never automatically be mistaken for a correct one. There is another question businesses need to ask.

What are your employees putting into AI?

Think about the information contained within a typical grievance or workplace investigation.

• Names.
• Health information.
• Witness statements.
• Allegations of harassment or discrimination.
• Salary information.
• Performance concerns.
• Confidential company information.
• Potentially commercially sensitive documents.

An employee copying an entire investigation report into an AI tool because they want it summarised creates a very different issue from asking AI to help them improve the wording of a generic email.

Businesses therefore need to think beyond simply deciding whether employees are "allowed to use ChatGPT".

The better questions are:

• What can employees use AI for?
• What information can they put into it?
• What information should never be entered?
• When must an AI generated answer be independently checked?
• Can managers use it when dealing with grievances, disciplinaries and investigations?
• Should AI ever be asked to recommend an employment decision?

And most importantly, who is responsible when it gets something wrong?

Hancox gives us a useful answer to that last question. AI is a tool, not an escape from responsibility. The EAT did not conclude that AI has no place in legal proceedings.

Quite the opposite.

The judgment recognised its potential benefits while setting out the responsibilities that come with using it.

And there is an important distinction here. Using AI to help you think is not the same as allowing AI to think for you.

• Use it to organise information.
• Use it to help you understand something.
• Use it to generate questions.
• Use it to produce a first draft.
• Use it as a starting point for further research.

But don't switch your judgement off simply because the answer appeared on screen in three seconds.

Mr Hancox submitted 300 pages of material generated using ChatGPT. The EAT found that the document did not assist it in preparing for the hearing and left it with what the judge described as an "unmanageable document".

The technology wasn't ultimately responsible for what was submitted. The person using it was.

And that may be the most important lesson from Hancox for employers.

Artificial intelligence shouldn't make you unintelligent.

The faster AI becomes, the more important human judgement may become.