UK Upper Tribunal: immigration adviser gives Google AI Mode as his best guess for a non-existent case in grounds he drafted; he self-reported to regulators
In the first of two matters decided together in [2026] UKUT 81 (IAC), promulgated on 17 November 2025, the Upper Tribunal (Immigration and Asylum Chamber) examined grounds of appeal dated 17 March 2025 that cited a non-existent Court of Appeal case under the citation of an unrelated equal-pay case. The adviser who drafted them first denied using AI, then told the tribunal his best guess was that he had inadvertently used Google's 'AI Mode', and the tribunal recorded that he had concluded it was probable the case came from inadvertent AI use; the panel itself got Google AI to produce varying benches for the fictitious case. He reported himself to the Immigration Advice Authority and the Solicitors Regulation Authority, and the tribunal said it would otherwise have referred him. He also told the tribunal that he had put client emails and uploaded Home Office decision letters into ChatGPT to summarise them, which he called a data breach he would disclose to the clients; the tribunal said such uploads breach client confidentiality and waive privilege.
- AI system
- Google AI Mode (reported)
- Occurred
- 17 Mar 2025
- Reported
- 17 November 2025
- Event location
- United Kingdom
- What the AI did
- Communicated with the person
- Reported harm
- Professional Harm
- Whose AI use
- Their own AI use
- Setting
- Justice · Work
- Evidence
- AI involvement reported · Causal attribution supported · 1 source
- 4 claims: 2 documented, 2 reported. 4 open questions
- People reported harmed
- 1 person
AI system as recorded: Google search 'AI Mode', identified by the adviser as his best guess for the source of the false case
What Happened
The false case. The Upper Tribunal issued a show cause notice while considering an application for permission to appeal, asking the adviser's firm to explain why a case cited in the grounds of appeal (Horleston v SSHD [2007] EWCA Civ 654) could not be found and why its citation belonged to an unrelated equal-pay case, and to state whether an AI large language model such as ChatGPT had drafted the grounds. The adviser said he drafted the grounds of 17 March 2025 and at first stated that no AI large language model was used, attributing the error to human error.
The explanation given to the tribunal. In a witness statement of 10 October 2025 he said he could not dismiss the possibility that the case was an AI creation, as there was no other explanation. At the hearing his best guess was that he had inadvertently used the 'AI Mode' of a Google search. The tribunal recorded that he admitted including one fake case and had himself concluded that it was probable this happened through inadvertent use of an AI large language model. The panel found no reported case of that name, and by asking Google AI the same question in different ways it elicited different compositions of the Court of Appeal bench said to have decided the fictitious case.
Consequence for the adviser. He reported himself to the IAA and the SRA on 14 October 2025. The tribunal made no referral because of those self-reports and said it would have referred him otherwise.
Client documents in ChatGPT (context: a separate, undated matter not counted in this case). Questioned by the panel, he said he had put client emails explaining Home Office decisions into ChatGPT to improve them and had uploaded Home Office decision letters to it to summarise them for clients; he said he now realised this was a data breach and would inform the clients, the IAA and the SRA. The tribunal stated that putting such letters into an open-source AI tool such as ChatGPT places the information in the public domain, breaches client confidentiality and waives legal privilege.
What is not established. The tribunal's finding on the source of the false case rests on the adviser's own conclusion and best guess; no search history or AI output from the drafting is described. The effect of the false case on the appellant's application, the number of clients whose documents were uploaded and whether they were told are not stated.
Reported harm
According to the Upper Tribunal's judgment, an immigration adviser placed a non-existent case in appeal grounds, which the tribunal found he had concluded probably came from inadvertent AI use (his best guess: Google AI Mode); he self-reported to the IAA and the SRA, and the tribunal said it would otherwise have referred him.
Outcome
OngoingThe adviser self-reported to the Immigration Advice Authority and the Solicitors Regulation Authority on 14 October 2025; the tribunal made no referral because of the self-reports and said it would otherwise have referred him. He said his firm would stop using AI tools and that he would inform the affected clients, the IAA and the SRA about the ChatGPT uploads. Any regulatory outcome is not reported in the judgment.
What remains unknown
- Whether the false case actually came from Google AI Mode; the judgment records the adviser's best guess and conclusion, with no search history or output
- What effect, if any, the false case had on the appellant's permission application
- How many clients' emails or decision letters were put into ChatGPT and whether they were informed
- The outcome of the IAA and SRA self-reports
What the evidence supports
AI involvement: reported. The judgment records the adviser's best guess that the non-existent case came from his inadvertent use of Google's 'AI Mode', and finds that he concluded it was probable the case was included because of inadvertent use of an AI large language model; the panel reproduced Google AI generating different benches for the same fictitious case. The output reached him through his own search (communicated_with) and went into the grounds that led to the show cause notice and his self-reports. This rests on his account as recorded by the tribunal; he first denied AI use. His separate account of putting client emails and Home Office decision letters into ChatGPT is described as context and is not part of this case's harm.
4 claims: 2 documented, 2 reported. What the statuses mean
Documented Grounds of appeal dated 17 March 2025, drafted by the adviser, cited a case (Horleston v SSHD [2007] EWCA Civ 654) that the Upper Tribunal could not find, under the citation of an unrelated case; the tribunal issued a show cause notice.
Causal attribution. Established by the judgment's own record of the show cause notice and the adviser's response.
- caselaw.nationalarchives.gov.uk(opens in new tab) supports · English
'he was the person responsible for the case and drafting of the grounds of 17th March 2025 in this matter.'; 'provide an explanation for the fact that the case Horleston v SSHD [2007] EWCA Civ 654 cited in the grounds of appeal to the Upper Tribunal is not available on BAILII'
Reported The adviser first denied using an AI large language model, later said he could not dismiss that the case was an AI creation, and at the hearing gave as his best guess that he had inadvertently used Google's 'AI Mode'; the tribunal found he had concluded it was probable the fake case came from inadvertent AI use, and the panel elicited varying benches for the fictitious case from Google AI.
Causal attribution. The tribunal documents what the adviser said and its own test of Google AI; the source of the case in the drafting rests on his conclusion and best guess, so the claim is reported.
- caselaw.nationalarchives.gov.uk(opens in new tab) supports · English
'We confirm unequivocally that no AI large language model (such as ChatGPT) was used in drafting the grounds of appeal.'; 'In absence of an explanation and with how AI operates, I cannot dismiss the fact that the case was an AI creation as there is no other explanation.'; 'his best guess was that he had inadvertently used the “AI Mode” of a Google search by clicking on this on the bar under the search topic.'; 'concluded himself that it was probable that this had happened because of the inadvertent use of an AI large language model'; 'As far as the Panel could establish there had never been a reported case called Horleston but Google AI nevertheless can, depending on the question posed, produce information such as panels of Court of Appeal judges who sat on this non-existent case'; 'By asking Google AI the same question in slightly different ways, we were able to elicit various different compositions of the bench which decided that fictitious case.'
Documented The adviser reported himself to the IAA and the SRA for having unknowingly inserted a non-existent case, and the tribunal made no referral for that reason, stating it would otherwise have referred him.
Causal attribution. Established by the judgment's findings and its stated decision on referral.
- caselaw.nationalarchives.gov.uk(opens in new tab) supports · English
'reported himself to the IAA and to the SRA for having unknowingly inserted a non-existent case in grounds of appeal'; 'has already done this and there is no need for us to do so. If he had not done this we wish to make it clear that we would have made a referral'
Reported The adviser told the tribunal he had put client emails explaining Home Office decisions into ChatGPT and uploaded Home Office decision letters to it to summarise them, called this a data breach he would disclose to the clients, the IAA and the SRA; the tribunal said such uploads breach client confidentiality and waive legal privilege.
Causal attribution. The uploads rest on the adviser's own account as recorded by the tribunal; the tribunal's statement on confidentiality is its own observation. The clients' number and whether they were told are not stated.
- caselaw.nationalarchives.gov.uk(opens in new tab) supports · English
'he had put client emails he had drafted explaining Home Office decisions into ChatGPT to try to improve them and he had uploaded Home Office decision letters to this platform to summarise them for clients.'; 'He informed us that he now realises that this is a data breach and will inform the clients that he has done this, as well as the IAA and the SRA.'; 'to put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege'
Sources
1 source inspected. Sources that repeat one account do not corroborate each other.
- Find Case Law, [2026] UKUT 81 (IAC), UK v SSHD (AI hallucinations; supervision; Hamid), promulgated 17 November 2025(opens in new tab)
s1 · caselaw.nationalarchives.gov.uk · Court record · English · Inspected · 17 November 2025 · Primary
How the sources were read, and where the events happened
Full judgment text read on Find Case Law (HTTP 200); no translation. Applies to s1.
Event countries: United Kingdom. Affected-person countries: Unknown. Court countries: United Kingdom.
Event: the grounds were filed with the First-tier and Upper Tribunal (Immigration and Asylum Chamber), heard at Field House, and the adviser practises under the UK regulators IAA and SRA (judgment). Court: the Upper Tribunal. The adviser's and clients' countries of residence are not stated.
Reviewed for publication 2026-10-11: Published from the full Upper Tribunal judgment: it documents the false case in grounds the adviser drafted and his self-reports to two regulators, and records his conclusion that the case probably came from inadvertent AI use and his best guess that the tool was Google AI Mode, which the panel's own Google AI test illustrates. The AI source is attributed to his account; his separate account of uploading client documents to ChatGPT is context. The adviser is described by role; the anonymised appellant is not counted.
People described
An immigration adviser (a level 3 IAA-accredited adviser who is also a solicitor) whose grounds of appeal contained an AI-sourced non-existent case and who self-reported to his regulators
People reported harmed in this case
1 person
1 AI participant · 0 other people harmed
One counted person: the adviser, who used the AI and faced the regulatory consequence. His clients whose emails and decision letters he said he put into ChatGPT belong to a separate, undated matter described as context and are not counted. The appellant in the matter is not counted because no consequence of the false case for the appeal is stated.
Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.
Tags
Cite this case
Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.
APA
NOPE. (2026). UK Upper Tribunal: immigration adviser gives Google AI Mode as his best guess for a non-existent case in grounds he drafted; he self-reported to regulators. AI incidents. https://nope.net/incidents/2025-uk-upper-tribunal-immigration-adviser-google-ai-mode-fake-case-in-appeal-grounds-regulator-self-report
BibTeX
@misc{2025_uk_upper_tribunal_immigration_adviser_google_ai_mode_fake_case_in_appeal_grounds_regulator_self_report,
title = {UK Upper Tribunal: immigration adviser gives Google AI Mode as his best guess for a non-existent case in grounds he drafted; he self-reported to regulators},
author = {NOPE},
year = {2026},
howpublished = {AI incidents},
url = {https://nope.net/incidents/2025-uk-upper-tribunal-immigration-adviser-google-ai-mode-fake-case-in-appeal-grounds-regulator-self-report}
} Related cases
SDT strikes solicitor Abhishek Kumar off the Register of Foreign Lawyers after AI-generated false citations in his SRA defence
On 25 August 2026 the Solicitors Disciplinary Tribunal struck Abhishek Kumar off the Register of Foreign Lawyers after finding proved that his 12 March 2026 Answer to the SRA's Rule 12 Statement contained misleading quotations and citations produced with generative AI — including a non-existent 'SRA v Chan [2020] EWHC 1502' and a miscited 'SRA v James, MacGregor & Naylor [2018] EWCA Civ 1420' that is actually an intellectual-property case — and that his 9 April 2026 email admitting AI use was itself AI-drafted with further errors. The tribunal said it would have struck him off on that allegation alone; the parallel ground was his January 2024 conviction under s.21 of the Immigration, Asylum and Nationality Act 2006. This is the SDT's first case on a lawyer's use of AI, per the SRA's counsel.
Jeju: police refer a middle-school student over secretly filming teachers and making sexual deepfake composites; teachers dispute victim count and legal aid
Jeju Dongbu police announced on 23 September 2026 that a student at a middle school on Jeju had been referred to prosecutors without detention on 17 September for violating the Sexual Violence Punishment Act (filming with a camera or similar device). The student's phone was voluntarily submitted by his guardians and forensically examined; the examination found footage relating to some teachers and some sexual composites, and the student admitted the conduct. News1 reported that the student was booked for filming a teacher's body with his phone during a school field trip around mid-July 2026, that the offending had continued for several months with multiple victims, and that police found no indication the composites had been distributed. Police told Seoul Shinmun there were more victim teachers than the one first reported but withheld the number to avoid secondary harm, and that no student at the school was a victim; the Jeju teachers' union questioned whether the full scope of harm had been established. According to the union, two months on the victim teachers were receiving hospital treatment and counselling while teaching daily, gave police statements on 14 September without a lawyer, attended the regional teachers' rights protection committee on 17 September without institutional support, wrote the criminal complaint themselves, and were offered 3.3 million won in legal fees for all of them together, payable after the case ends; two requests to meet the superintendent were not accepted. On 29 September one of the victim teachers gave a press interview at the Jeju Teachers' Union office. She said that early in the investigation she had seen on the student's phone material showing other teachers and current students, had made a list of the people and types of harm, and that the material shown to her at the police victim examination was considerably less than what she had seen; Yonhap-chain outlets quote her as saying the students she recognised numbered 10 to 20 and that a graduate had suffered worse deepfakes. She said investigators asked the school only to identify the perpetrator and never requested material to identify the students in the footage, and that the school sent parents a letter premised on there being no student victims. Police replied that phone forensics had been thorough, that every act beyond social norms had been identified and proven, and that one student victim had been confirmed, a student at a different school. The teachers were notified on 22 September that the case had gone to the prosecution and on 24 September that it had been sent to the court's juvenile division as a juvenile protection case, before they had submitted further evidence and a victim opinion. The teacher said the original complaint had also asked police to examine whether physical contact in 2025 was forcible molestation and whether repeated approaching and waiting in 2026 was stalking, and Halla Ilbo reports that such conduct is said to date from 2025. Yonhap-chain reports describe the charged victims as five teachers. The same day the education office briefed on legal-aid rules (up to 6.6 million won per instance when a teacher is sued, 3.3 million won per case when a teacher files a complaint; office lawyers cannot act as a teacher's private counsel), said it had linked the teachers to the national Digital Sexual Crime Victim Support Center, announced an integrated support system through a hotline, and its section head said the office was sorry that early support had not been felt. The KTU Jeju branch (28 September) and the Jeju Teachers' Union (14 and 29 September) demand a victim-support system, per-teacher legal fees paid in advance and a dedicated support team. On 6 October 2026 Seogwipo Broadcasting reported, citing people in education and a tipster, that the education authority had finally rejected the victim teachers' application for lawyer fees for an additional criminal complaint, saying in a written opinion that with a police investigation already under way a further complaint could hardly stop or prevent the infringement of their teaching, so the support requirements were not met; critics contrast this with full legal-cost support in earlier sexual-offence cases.
Former federal prosecutor in North Carolina publicly reprimanded after admitting that a brief with fabricated quotations was drafted with AI; he lost his post
In December 2025 an Assistant U.S. Attorney in the Eastern District of North Carolina filed a response brief for the Department of Defense in a self-represented veteran's lawsuit over health-insurance coverage. The plaintiff identified fabricated quotations and misdescribed case holdings in it. The attorney first attributed the errors to an unfinalized draft filed by mistake. On 2 March 2026 Magistrate Judge Robert T. Numbers II ordered him to show cause why he should not be sanctioned. At the 10 March hearing the attorney, Rudy Renfer, said under oath that he had accidentally overwritten his draft, panicked and used artificial intelligence to draft the brief, and had not verified the cases it cited; he said he had resigned after 17 years in the office, and Courthouse News reported he was terminated the next day. On 28 April 2026 the court issued a public reprimand instead of a fine, writing that his reputation was in tatters and that his job loss already exceeded the usual fines.
Fee suit against the City of Aberdeen (N.D. Miss.): four attorneys on both sides sanctioned after AI tools produced fabricated case citations; trial cancelled and both litigants left without counsel
In a fee dispute between a Louisiana attorney and the City of Aberdeen, Mississippi, briefs filed for both sides in late 2025 cited six cases that do not exist. The attorneys admitted the citations came from unverified AI use: the plaintiff's out-of-state counsel, Kathleen M. Wilson, drafted her filing with an AI drafting program called 'First Drafts', and the City's out-of-state counsel, Kathryn Y. Williams, used an in-house AI legal research tool. Senior Judge Sharion Aycock stayed the case and cancelled the March 2026 trial, then on 8 June 2026 revoked both attorneys' pro hac vice admissions, barred them from the district for two years and fined them $2,500 and $3,500, and disqualified and fined the two local counsel who had signed the filings. Both litigants were left without counsel and given 60 days to find new representation.
If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.