Naughty baby! UK parties admonished for AI use in court

Tali Blecher

Tali Blecher

We’ve seen a few legal decisions in which one side is chastised for their uncritical reliance on AI. This is the first case we’ve seen where both parties get torn to shreds.

What happened?

The case (In the matter of Trade Mark Application no. 3887112 in the name of Pro Health Solutions Ltd) was an appeal from a decision of the UK Registrar of Trade Marks. It was a fight between two parties about whether one party could register the trade mark PROHEALTH.

The early stages of the appeal required both sides to prepare and file certain documents. Those documents contained the following issues:

  • Party A filed quotes from earlier court decisions. But the quotes weren’t found in the judgments.
  • Party A also filed a skeleton of arguments containing summaries of certain prior court decisions. Those summaries were manifestly incorrect.
  • Party B (represented by an attorney) filed a skeleton of arguments with reference to three cases that were irrelevant and did not stand for the propositions the attorney said they did.

The suspicious decisionmaker (known in the UK as the Appointed Person) raised his concerns with both parties at the hearing.

Party A quickly admitted that he used ChatGPT to draft the appeal documents. He apologised unreservedly. He got off fairly lightly.

Party B fared worse. When the Appointed Person asked Party B’s attorney to identify the parts of the cases that supported his arguments, he responded, “I cannot actually remember that now, to be honest with you” before attempting to lead the Appointed Person down the garden path. You can guess the rest.

The Appointed Person criticised both parties for their over-reliance on AI, but reserved special condemnation for Party B’s attorney who waltzed dangerously close to precipice of misleading the court. The Appointed Person considered reporting the attorney to the UK’s IP Regulation Board but decided not to – only because of the slim possibility that the attorney was not out-and-out lying; that he had read the cases he cited but grossly misinterpreted them.

AI is an adult baby

As AI-powered legal tools develop and mature, it’s only going to get easier and more tempting for parties to incorporate LLM-generated content into their legal documents. Who could blame them? AI, when used correctly, is a game-changing, quality-improving time saver.

So here are some things to keep in mind Every. Single. Time you use AI for professional purposes:

  • Legal representatives: you can get publicly embarrassed (at best) or disbarred (at worst) for relying uncritically on AI.
  • Unrepresented litigants: you may have the winning case, but unskilled use of AI could result in a cost penalty (at best) or dismissal of your case (at worst). Check AI’s work. If you don’t know how to do that, here’s a bitter pill: you need a lawyer.
  • The standard advice to practitioners to scrutinise AI-generated work as though it was generated by a junior lawyer is unkind to juniors. A junior lawyer might misconstrue portions of a judgment, but they generally won’t pull citations from their nether regions. Nor will they generally assert things with total confidence if they harbour doubt in their own correctness.
  • Treat AI platforms as extremely sophisticated, psychotic adult babies. They charm you one minute, seducing you into unquestioning allegiance with their eloquence and logic, before throwing a runny turd (fabricated citation) your way. Beware.

More cuddles, less tantrums

As with raising a baby, our aims in using AI are to optimise for cuddles and against tantrums.

Cuddle-optimisation means using the adult baby for its strengths. These are things like summarising source material, improving the structure of pre-written content, extracting key data like dates, and suggesting edits to reduce word count.

Tantrum-prevention means being smart to the adult baby’s weaknesses. As at July 2025, I think it’s safe to advise against using AI to conduct primary legal research – unless you have the skills and patience to closely vet the work product.

AI’s greatest weakness is regard for the truth. Beware the adult baby.

Share Post

Share on facebook
Share on google
Share on twitter
Share on linkedin