TL;DR
The Solicitors Disciplinary Tribunal has handled its first AI hallucination case, and the answer to “what actually happens to you” is now on the record: erasure. Abhishek Kumar, a registered foreign lawyer, defended himself against disciplinary allegations using generative AI, produced quotations and citations that did not support what he claimed they supported, and then compounded it — his reply to that charge was itself AI-drafted and contained fresh fabrications.
How it unfolded
Kumar, 41, reached the tribunal by another route entirely. He had been convicted in 2024 of knowingly employing someone disqualified from work by immigration status — a client whose asylum claim had been withdrawn, paid cash in hand for administrative work at his firm. He received a 12-month community order and 150 hours of unpaid work.
The AI charge was bolted on afterwards. When the regulator’s counsel read his written answer, two of the cited authorities turned out not to stand for the propositions attached to them. Kumar admitted using generative AI and said he lacked the expertise to check its output. He argued this was incompetence and poor judgement under severe health difficulties, not dishonesty, and that the High Court’s ruling in Ayinde did not bind him because he was a registered foreign lawyer rather than a solicitor.
The tribunal rejected the distinction outright. He was regulated, and therefore bound by his profession’s standards. Citing authorities for things they do not say amounts to putting submissions that are not properly arguable — a breach in its own right.
Why the second use mattered more
The tribunal gave “very substantial weight” to his having reached for AI again to fix what AI had broken. It found he had started using the technology without any thought as to the consequences, and took no steps to check either his output or whether the tools were reliable. Culpability was assessed as very high, alongside a persistent lack of insight.
Looking forward
Two details matter for UK firms. The tribunal was explicit that either allegation alone would have produced the same sanction — so this is not a case where AI misuse merely topped up an existing penalty. And it drew a distinction worth noting: inventing a case outright may sit differently from getting a citation wrong. That leaves room for proportionality in future rulings. It arrives days after our coverage of lawyers’ hallucination fears rising even as adoption reached 94% — a gap between anxiety and verification practice that this ruling prices.