Solicitors Regulation Authority v Abhishek Kumar

The record

• TribunalSolicitors Disciplinary Tribunal, a statutory tribunal under the Solicitors Act 1974, not a court
• CaseNo. 12884-2026, Solicitors Regulation Authority Ltd v Abhishek Kumar
• PanelMr P Lewis, chair; Ms H Hasan; Mr D Kearney
• Heard29 July 2026. Judgment dated 25 August 2026, published on the Tribunal’s site and reported in the legal press on 4 September 2026
• AppearancesJoshua Hitchens of Outer Temple Chambers for the SRA. Mr Kumar represented himself
• StatusRegistered Foreign Lawyer, registered in 2018. He was not a solicitor, and he was not struck off the roll of solicitors
• SanctionStruck off the Register of Foreign Lawyers. No order as to costs, the SRA accepting that he was impecunious
• AllegationsTwo, both proved. The Tribunal recorded that it would have imposed the same sanction on either one standing alone
• ToolNot named anywhere in the judgment. The word “hallucination” does not appear in it either
• DishonestyNot alleged, on either allegation
• BreachesPrinciples 1, 2 and 5 of the SRA Principles 2019, and paragraphs 1.4 and 2.4 of the SRA Code of Conduct
• AuthoritiesSRA v Sharma [2010] EWHC 2022 (Admin) and SRA v James [2018] EWHC 2058 (Admin) on sanction. No exceptional circumstances found

The two allegations

Allegation 1, the original matter. On 19 August 2020 he knowingly employed an adult who was disqualified from employment by reason of immigration status. He was convicted on 29 January 2024 at the Central Criminal Court under section 21(1) and (2) of the Immigration, Asylum and Nationality Act 2006, and on 26 April 2024 sentenced to a twelve month community order with 150 hours of unpaid work. Proved.

Allegation 2, added because of what he did while defending himself. On 12 March 2026 he filed a written Answer to the SRA’s statement of case “which contained misleading quotations from authorities and citations in consequence of the Respondent’s use of artificial generative intelligence”. On 9 April 2026 he emailed the SRA admitting he had used generative AI to prepare that Answer, and that email itself contained further erroneous citations and quotations. On 13 April 2026 he accepted that the email had also been drafted using generative AI. Proved.

The authorities he cited “either: (a) Did not exist; and/or (b) Had incorrect citations; and/or (c) Misquoted the principles set out in the authorities; and/or (d) Did not contain the passages quoted from them; and/or (e) Did not support the propositions for which they were cited.”

What the Tribunal found

“The Tribunal determined the Respondent’s culpability to be very high. A lawyer is always responsible for the accuracy of his or her output.”

That is paragraph 28.11. The next paragraph sets out what he did not do:

“The Tribunal found that the Respondent had embarked on the use of AI without giving any proper thought to the consequences of so doing. He had failed to take proper steps to ensure his work output was accurate or to satisfy himself that the electronic tools he was seeking to use were reliable.”

The tool is not the offence. At paragraph 19.48: “Whilst AI was, in this case, the medium through which the material was produced, it was not necessary for the erroneous material to have been generated by or with AI. The same issue might arise using common search engines or if work was produced by another person for which the respondent solicitor had ultimate responsibility.” The Missouri Court of Appeals reached the same position independently four weeks later in State ex rel. Schnell v. Richardson.

The duty is not confined to court filings. The Tribunal recorded that it is a statutory tribunal rather than a court, and then held at 19.47 that “there was no distinction in principle as to whether the material relied upon by a lawyer was sent to a regulator, a disciplinary tribunal or court.” The material in this case was sent to the regulator.

The boundary, in the Tribunal’s own words. “Mistakes happen. Not every mistake will be a matter serious enough to engage the Regulator or this Tribunal.” Separately, at 8.2, “Typographical errors in case citations were not sufficiently serious to amount to misconduct.” What tipped this case was repetition: culpability was very high because the conduct was “repeated after errors had been pointed out and explained”, and the Tribunal gave “very substantial weight to the repeated nature” of it.

It rejected the argument that the Ayinde expectations did not apply to him because he was a Registered Foreign Lawyer and not a solicitor. It rejected his description of his own conduct as “technological negligence” and his claim that he “mistakenly believed” AI was a “lawful drafting aid”, calling that “a persistent lack of insight”.

Two separate failures, and what that finding is not

Read paragraph 28.12 slowly. The Tribunal faulted him twice over, and the two faults are different. The first is the familiar one: he did not take proper steps to ensure his output was accurate. The second is not familiar, and it is the part of this judgment that has no counterpart elsewhere in this record: he did not satisfy himself that the tools he proposed to use were reliable.

Read it for what it is. This is a finding by a disciplinary tribunal about what this respondent failed to do on these facts, weighing his culpability. It is not a statute, it is not a practice direction, and it does not create a general, freestanding duty on every lawyer everywhere to audit a vendor before opening it. No instrument in this record imposes such a duty in those terms. What the finding does show is the direction: when a regulator asks how the work was done, “the software produced it” was not treated as an answer, and neither was silence about whether the software could be relied on.

Why it matters, and why it is the answer to Ayinde

The Tribunal said itself, at 8.3, why it treated the case as a landmark: “This was the first time that a lawyer’s use of artificial intelligence or large language models in legal proceedings had been litigated before the Tribunal.” It recorded that the High Court in Ayinde v Haringey LBC [2025] EWHC 1383 (Admin) “had invited the legal regulatory bodies to consider, as a matter of urgency, what further steps, beyond issuing guidance, needed to be taken to address this problem”, and that the SRA considered, and the Tribunal accepted, that this was “an issue of profound public importance”.

Ayinde asked the regulators a question in June 2025. This is what the answer looks like when it arrives: not new guidance, but a prosecution, a proved allegation and a name removed from a register.

The Tribunal set out the Ayinde passages it relied on, including the sources against which AI research is to be checked: the Government’s legislation database, the National Archives database of court judgments, the official Law Reports of the Incorporated Council of Law Reporting for England and Wales, and the databases of reputable legal publishers.

Sources

• SRA v Kumar, Case No. 12884-2026, the judgment, twenty-three pages, the copy read in full for this page
• The Tribunal’s own case record for 12884-2026
• Ayinde v Haringey LBC [2025] EWHC 1383 (Admin), the judgment that asked the regulators to act

Related

• The full sanctions record
• Ayinde v Haringey, the question this decision answers
• Bandla v SRA, the other England and Wales matter on this record
• Falkenstein v. Niles, a federal judge asking what steps counsel took to verify
• Verification is becoming a duty. Evidence of verification is the missing layer.

Verified 11 September 2026 against the judgment, read in full. Information, not legal advice, and not advice on English law.