Two people connected with an east London law firm have reported themselves to the Solicitors Regulation Authority after court documents generated with ChatGPT were found to contain fabricated cases, invented quotations, and false statements about what a judge had said. The case, revealed in a newly reported judgment by Upper Tribunal Judge Keith, has become a stark warning about the dangers of using artificial intelligence in legal work without proper oversight. The documents were filed by Sarker Solicitors in an immigration appeal. According to the judge, the grounds of appeal were “completely infected” by AI and “seriously misleading.” This was not a simple mistake. The AI did not just miscite a case; it invented cases, gave them names, attributed quotations to them, and even fabricated passages from a real decision by First-tier Tribunal Judge Balroop. The two individuals involved have voluntarily reported themselves to the SRA, the body responsible for regulating solicitors in England and Wales. That self-reporting is a significant step, but it does not undo the damage. The case shows how easily a tool that promises efficiency can undermine the integrity of the legal process. It also raises serious questions about supervision, confidentiality, and the duty of lawyers to check every word they put before a court. For the legal profession, this is a moment to pause and reflect on what it means to be a responsible lawyer in the age of AI.
The underlying appeal was brought by a Namibian national who had been refused asylum by the Home Office. Her first appeal was dismissed by First-tier Tribunal Judge Balroop. She then asked the Upper Tribunal for permission to appeal. Her solicitors filed grounds of appeal and later a skeleton argument, which is a written summary of legal arguments. Both documents were drafted by an unnamed paralegal using ChatGPT. Judge Keith said this was “one of the more extreme examples of the misuse of AI, rather than a mere mis-citation of particular cases.” He went on: “The misstatements extended not only to what the judge had supposedly said and to incorrect case citations, but also to assertions of legal propositions which were not supported by the cases cited and also misstatements as to the evidence. The matter could not have been more serious.” The Home Office, which was the opposing party in the asylum case, was even blunter. Its counsel argued that the firm’s misrepresentations were “at best negligent and at worst reckless.” In ordinary language, that means the government was accusing the solicitors of acting without due care, and possibly with a disregard for whether the information they were putting into the appeal was true. The judge did not go so far as to say the firm was dishonest, but his words were harsh enough.
What made the filings so dangerous was that the false material was not peripheral. The grounds of appeal quoted Judge Balroop’s decision in passages that were “pure invention.” These quotes were not simply inaccurate; they did not exist anywhere. Yet the solicitors used them as the foundation for one of their grounds of appeal, criticising Judge Balroop for statements she had never made. The grounds also cited fake cases and quoted paragraphs from those fake cases as if they were binding authority. When the Upper Tribunal granted permission to appeal, the solicitors were explicitly told that Judge Balroop had been wrongly misquoted. That warning should have prompted an immediate review. A careful solicitor would have gone back to the original judgment, checked every quotation, looked up every case, and corrected or withdrawn the grounds. Sarker Solicitors did none of this. They did not amend the grounds, and they did not address the many flaws. Instead, they filed and served a skeleton argument that was based on the same defective grounds. Judge Keith noted that the appeal had an “unhappy history” and illustrated the “considerable procedural difficulties” that can arise when AI is misused. He also made clear that the procedural irregularities were not caused by AI itself. They were caused by the solicitors’ misuse of AI. The tool was not responsible; the humans were.
The judge was also critical of the firm’s internal management. The paralegal who produced the documents was a junior member of staff. The supervising solicitor, named as Mr Sarker, “had not assumed any managerial responsibility” for the paralegal’s work. Judge Keith said the failures happened “because Sarker Solicitors did not check the accuracy of the grounds of appeal, did not consider the grant of permission and did not check the accuracy of the skeleton argument.” He added that both documents were produced by a paralegal using ChatGPT, “which itself raises significant concerns about client confidentiality and legal privilege.” This point matters beyond the immediate case. Solicitor-client privilege is one of the most important protections in the legal system. It allows clients to speak openly to their lawyers without fear that what they say will be used against them. If confidential information is typed into a public AI tool, that protection can be lost. The information may be stored by the technology company, accessed by other users, or used to train future models. The judge said that throughout the process, Mr Sarker’s supervision “did not begin to mitigate the risk of filing seriously misleading submissions.” That is a devastating assessment. It suggests that the supervisor was either absent, unaware, or unwilling to take responsibility. In the end, the judge declined to set aside permission to appeal, but he dismissed the appeal anyway. He also invited the Home Office to apply for a wasted costs order against Sarker Solicitors within 28 days. A wasted costs order is a court order requiring a legal representative to pay costs that were caused by their improper, negligent, or unreasonable conduct.
This case is not just about one firm. It is a warning to the entire legal profession. Artificial intelligence is already changing the way lawyers work. It can help with research, drafting, and summarising. It can reduce costs and speed up routine tasks. But AI has a well-known tendency to “hallucinate”—to invent facts, cases, and quotes that look completely real but are entirely false. A lawyer who uses AI must therefore act as a careful editor and guardian. Every citation must be checked against an original source. Every quote must be verified. Every legal proposition must be tested against the actual law. The lawyer cannot delegate this responsibility to a junior employee, and certainly cannot delegate it to a machine. The case also raises questions about how the SRA will respond. The fact that the two individuals reported themselves is important. The SRA is likely to investigate whether they breached its principles, which include acting with integrity, maintaining public trust, and providing a proper standard of service. Possible outcomes range from fines and conditions on practice to more serious sanctions. The case may also prompt the SRA and the courts to issue new guidance on the use of AI in litigation. For now, the message is simple: technology is a tool, not a substitute for professional judgment. If you do not understand what the machine has produced, and if you cannot verify it, you have no business putting it before a court.
Behind the legal doctrine and regulatory talk, there are real human stakes. The appellant in the case was a woman from Namibia who had sought asylum in the United Kingdom. She was not a corporation or a test case; she was a person trying to escape danger and build a life. Her appeal was dismissed, in part because the legal arguments presented on her behalf were not credible. She was badly served by the people she trusted to represent her. It is easy to imagine the paralegal, working late and under pressure, asking an AI chatbot for help with legal drafting. The chatbot produced something that looked polished and persuasive. It had citations, quotations, and legal phrases. It looked like the work of a qualified lawyer. But it was a mirage. The paralegal probably did not intend to mislead anyone; the danger is precisely that the technology made the deception easy and invisible. Mr Sarker, as the solicitor responsible for the firm’s work, failed in his duty to supervise. The two individuals who have reported themselves to the SRA have at least taken the first step toward accountability. But this story is a painful reminder that in law, the details matter. A wrong case citation can waste time and money. A fake quote can destroy trust. A careless filing can cost a person their freedom, their home, or their life. The lesson is not to reject AI. It is to use it with humility, caution, and rigorous human oversight. No machine can replace the solicitor’s duty to read, understand, and stand behind every word. Justice depends on that duty. When lawyers forget it, the whole system suffers.