Anyone who has handled a grievance or tribunal claim over the last year or so will recognise the signs. The 40-page grievance that lands at 11pm. The polished legal phrasing that doesn’t quite sound like the person who sent it. The confident references to cases that, on closer inspection, don’t say what they’re supposed to say (or don’t exist at all…). Generative AI has handed every employee a free, tireless and extremely self-assured legal assistant. For HR teams, it has become one of the biggest drains on time in employment disputes today.
So we were very interested to read Hancox v Sutherland & Others [2026] EAT 139, handed down last week. As far as we’re aware, it’s the first appellate employment decision to take a hard look at AI use by a litigant. Helpfully for employers, it contains some very quotable principles.
Not so much a skeleton
The facts are a little unusual. Mr Hancox was dismissed by the National Farmers’ Union during his probationary period in 2020. He brought a series of claims, including whistleblowing, unfair dismissal and disability discrimination. His claim against four individual colleagues was struck out in 2021, after the tribunal found that a sustained LinkedIn campaign had made a fair trial impossible. That campaign ran to over 145 posts, some attaching documents disclosed in the litigation.
On appeal, and acting in person, Mr Hancox filed a “skeleton argument” two days before the preliminary hearing. It was 300 pages long and contained almost 132,000 words. For context, the EAT’s Practice Direction suggests 5 to 15 pages, with 20 as the upper limit. The judge, Marcus Pilgerstorfer KC, didn’t mince his words: “The word ‘skeleton’ is entirely inapposite for that document.“
Mr Hancox explained he had used ChatGPT because he needed to prepare quickly. He accepted at the hearing that much of the document lacked credibility, and it wasn’t relied on.
The part that really caught the judge’s attention was the covering email. It invited the respondents to identify any factual inaccuracies, incorrect quotations or mistaken references in writing by 4pm the following day. It also said that complaining the document was simply too long wouldn’t count! In effect, the AI did the drafting and the other side was expected to do the checking. The EAT made clear that simply isn’t how this works.
What the EAT said
The judgment is careful not to demonise AI. It recognises the genuine value AI tools can have for people without access to legal advice. A footnote even notes that some judges find AI-assisted submissions from litigants in person easier to digest. But it sets out some clear principles, and for employers several of them are gold dust:
Personal responsibility applies to everyone: Litigants in person aren’t bound by lawyers’ professional duties. But the EAT held that the principles of personal responsibility and accuracy apply to anyone who submits documents to a tribunal. At a minimum, AI-assisted documents must:
- comply with procedural rules;
- be checked for accuracy as thoroughly as the author reasonably can; and
- stick to relevant points, focusing on the best arguments and avoiding repetition.
Being unrepresented generally won’t justify a lower standard of compliance.
You can’t outsource the checking: In the judge’s words: “Simply submitting the product of generative AI to a court or tribunal, or placing the onus to check a document onto an opponent, is not acceptable.” Where checks look inadequate, judicial enquiries and potential sanctions are likely to follow.
The rules aren’t optional: “Compliance with the EAT’s procedural requirements is mandatory, not optional.” Another gem of a quote from the judge. The employment tribunals and the EAT have strict rules and a wide toolkit. Strike out and costs are therefore reasonable considerations in cases of excessive and inappropriate AI usage.
AI can undermine credibility: The judgment notes that AI use may make a tribunal more circumspect about how much weight to give a document or evidence, including AI-assisted witness statements. Equally, blaming false information on the AI will be met with careful scrutiny.
Resources and proportionality count: The EAT emphasised the limited resources tribunals face and the need to consider what is appropriate and proportionate when submitting materials and correspondence. They ordered that all future applications by Mr Hancox must be made in a prescribed form and set out “in one go” rather than in dribs and drabs. They also warned that email restrictions may follow if the excessive correspondence continued.
This problem starts well before the tribunal
The judge noted that AI is being used “with increasing frequency” in the EAT. He also pointed to recent Presidential Guidance observing that interim relief applications now often indicate AI use. That tallies entirely with what we’re seeing, except that in our experience the problem starts much earlier.
AI-assisted grievances are now a weekly occurrence across our client base. That goes for a runner on a production, a QA tester at a games studio, or a trader at a betting operator. They tend to share exactly the features the EAT criticised:
- they are long and repetitive;
- they list multiple overlapping complaints;
- they are heavy on legal labels and light on facts; and
- they often come with an expectation that every point is investigated in full, and quickly.
Many are followed by a steady drip of supplementary emails, each purporting to add something new (while not really doing that at all).
The difficulty is that most grievance policies were written for a world where a grievance ran to a page or two and the employee wrote it themselves. They simply weren’t built for this.
Back to basics
Our team is tackling this head on and has developed a new methodology for dealing with AI-generated grievances, complaints and claims. Its starting point is not technology but the grievance policy itself.
It’s striking how many of the tools the EAT reached for in Hancox that can be built into a well-drafted policy from day one. These can include considerations such as dictating a prescribed format, requiring complaints to be set out in one go, providing word and correspondence limits and (dare we say it) prohibiting AI usage altogether!
Our aim is not to discourage employees from raising complaints. Genuine complaints still need to be heard and dealt with properly and our client base fully understands that. The aim is to keep the process focused on the real issues rather than burying it under volume and preventing exorbitant costs from being incurred. Hancox is a welcome signal that tribunals and the judiciary see it the same way.
Final word (not written by ChatGPT, promise)
AI isn’t going anywhere and prohibiting its usage is likely a fruitless endeavour in most cases. But Hancox makes the point neatly: using AI doesn’t dilute responsibility for what you put your name to. For employers, the lesson is to be ready before the 300-page document lands, not to rewrite the rulebook mid-dispute.
If you’d like to talk to us about our AI grievance methodology or would like a review of your grievance policy, our specialists are here to help.