Law firm's ChatGPT-drafted appeal quoted a judge saying things he never said

A law firm in east London used ChatGPT to write court papers for an asylum appeal. The papers quoted a judge saying things he never said. They also cited fake cases. The Upper Tribunal has now set out what went wrong, in a decision issued on 21 September 2026 and published on 6 October.

This is not a story about a chatbot going rogue on its own. The tribunal was clear about that. "It would be misleading to describe them as having been caused by AI, rather than by the solicitors' misuse of AI," the decision says.

The firm, Sarker Solicitors, acted for a woman appealing against the refusal of her asylum claim. The tribunal made an anonymity order, so she cannot be named.

To take the case further, the firm drafted grounds of appeal. That is the document that says what the first judge got wrong. It quoted two sentences from the first judge's decision. The tribunal called this "pure invention", not found anywhere in that decision. Yet part of the appeal was built on criticising those words.

The woman's own barrister later accepted that the grounds were "completely infected by the use of AI". He listed fake cases, made-up passages from real cases, and made-up quotes from the first decision.

The warning signs came early. When another judge gave permission to appeal, he noted that the grounds misquoted the first judge in numerous places.

The firm did not fix them. Instead, it filed a skeleton argument, a written summary of the case, built on the same flawed grounds. Only when the Home Office's representative asked about citation errors, the day before a hearing, did the firm accept that AI had been used to produce both documents. That hearing had to be adjourned.

The tribunal said both documents "were produced by a junior member of staff, namely a paralegal, using ChatGPT". That, it said, "itself raises significant concerns about client confidentiality and legal privilege". It added that the supervision given by the solicitor in charge, Mr Sarker, "did not begin to mitigate the risk of filing seriously misleading submissions".

Two people involved have reported themselves to the Solicitors Regulation Authority, the decision says. The tribunal said nothing more about their conduct, because that is a matter for the regulator. No finding has been made.

The appeal itself failed. But the tribunal did not throw it out because of the AI. It let one rewritten ground go ahead, then decided the first judge had made no legal error.

Money is still an open question. The tribunal said "no decision has been taken on costs". It gave the Home Office 28 days to apply for wasted costs against the firm, or to say it will not pursue them. Nothing has been ordered.

The Law Gazette reported the case on 8 October.

Why does it matter? Lawyers can use tools. They must not mislead a court. Here, nobody checked the quotes against the very decision being appealed, even after a judge pointed out the problem. Much of the hearing went on issues that, the tribunal said, only arose because of the firm's inappropriate use of AI.