AI Hallucination Cases

Allegations that these are made-up quotes involving AI were aired in various blogs and online comments (see, e.g., here and here).

The tribunal eventually issued a correction, and then claimed that it was not AI but the result of "an exchange of correspondence between [the judge] and a judicial colleague" - which raises more questions than it answers.

The judge tested Microsoft M365 Copilot and found it produced fabricated Court of Appeal citations for Qdime and an incorrect citation for Gala Unity. The Tribunal concluded Mr Gurvits used AI-generated research without adequate verification; the AI outputs were rejected as unreliable. The Tribunal criticised the conduct but did not impose professional disciplinary sanctions; remedial orders in the case related to service charge and fee reimbursement.

The Claimant used AI to generate case citations in his pleadings; Tribunal found 46 inaccurate or misleading citations (9 wholly fictitious, 37 misrepresentations of real cases) and concluded the conduct was reckless and unreasonable, justifying a costs order.

The grounds of appeal contained at least two non-existent authorities. The judge concluded the false citations likely arose from unchecked generative-AI drafting and directed the solicitor to show cause why conduct should not be referred to the SRA.

Underlying judgment (in which the hallucination had been observed) is here.

The Father, acting pro se, relied in written submissions on numerous authorities (particularly about ASD) that HHJ Bailey identified as not genuine and apparently AI-generated; the court treated this as part of poor litigation conduct and awarded costs against him.

Claimant used Chat GPT to assist in drafting his grounds of appeal and skeleton argument. The documents included fabricated citations and misrepresented case summaries. Claimant admitted to using Chat GPT and apologized for the errors.

Compounding matters, the court suspected that the respondent had also used AI, since the cases cited in the Counsel's skeleton, though extant, did not support any of the propositions made - and Counsel was unable to explain how they got there.

This judgment, delivered on 6 June 2025 by the Divisional Court of the King's Bench Division, addresses two cases referred under the court's Hamid jurisdiction, which concerns the court's power to enforce duties lawyers owe to the court. Both cases involve lawyers submitting written arguments or evidence containing false information, specifically non-existent case citations, generated through the use of artificial intelligence without proper verification.

The Court used this opportunity to issue broader guidance on the use of AI in legal practice, raising concerns about the competence, training, and supervision of lawyers.

AI Use

Bandla denied using AI, claiming instead to have relied on Google searches to locate “supportive” case law. He admitted that he did not verify any of the citations and never checked them against official sources. The court found this unacceptable, particularly from someone formerly admitted as a solicitor.

Hallucination Details

Bandla’s submissions cited at least 27 cases which the Solicitors Regulation Authority (SRA) could not locate.

Bandla maintained summaries and quotations from these cases in formal submissions. When pressed in court, he admitted having never read the judgments, let alone verified their existence.

Ruling/Sanction

The High Court refused the application for an extension of time, finding Bandla’s explanations inconsistent and unreliable. The court independently struck out the appeal on grounds of abuse of process due to the submission of fake authority. It imposed indemnity costs of £24,727.20. The judge emphasized that even after being alerted to the fictitious nature of the cases, Bandla neither withdrew nor corrected them.

Key Judicial Reasoning

The court found Bandla’s conduct deeply troubling, noting his previous experience as a solicitor and his professed commitment to legal standards. It held that the deliberate or grossly negligent inclusion of fake case law—especially in an attempt to challenge a disciplinary disbarment—was an abuse requiring strong institutional response.

AI Use

Dr. Wright, after beginning to represent himself, repeatedly used AI engines (such as ChatGPT or similar) to generate legal documents. These documents were characterized by the court as "highly verbose and repetitious" and full of "legal nonsense". This use of AI contributed to filings containing numerous false references to authority and misrepresentations of existing law.

Hallucination Details

While the core issue in Dr. Wright's litigation was his fundamental dishonesty (claiming to be Satoshi Nakamoto based on "lies and ... elaborately forged documents" ), the use of AI introduced specific problems. His appeal documents, bearing signs of AI creation, contained "numerous false references to authority". His later submissions also involved "citation of non-existent authorities". This AI-driven production of flawed legal arguments formed part of his broader pattern of disrespect for court rules and process.

Ruling/Sanction

Mr Justice Mellor granted a General Civil Restraint Order (GCRO) against Dr. Wright for a three-year period. He found that an Extended CRO (ECRO) would be insufficient given the scope and persistence of Dr. Wright's abusive litigation. The court also referred Dr. Wright's conduct to the Attorney General for consideration of a civil proceedings order under s.42 of the Senior Courts Act 1981. Dr. Wright was ordered to pay the applicants' costs for the CRO application, summarily assessed at £100,000.

Key Judicial Reasoning

The court found "overwhelming" evidence that Dr. Wright had persistently brought claims that were Totally Without Merit (TWM), numbering far more than the required threshold. This conduct involved extensive lies and forgeries across multiple jurisdictions and targeted individuals who often lacked the resources to defend themselves. The judge concluded there was a "very significant risk" that Dr. Wright would continue this abusive conduct unless restrained. The court noted his consistent contempt for court rules and processes, including his perjury, forgery, breach of orders, and flawed submissions (including those using AI). A GCRO was deemed just and proportionate to protect both potential future litigants and the finite resources of the court system