AI hallucinations reach the Solicitors Disciplinary Tribunal

Last month, the Solicitors Regulation Authority (SRA) issued a warning notice identifying two areas of AI misuse: the submission of AI-generated false case citations to courts and breaches of client confidentiality (and the impact on legal privilege) through the use of public AI tools. The notice stressed the need for solicitors to check all information put before the courts, supervise work undertaken for clients, and remain accountable for everything produced using AI.

Following on from this, the Solicitors Disciplinary Tribunal (SDT) has now, for the first time, addressed the use of AI-generated false citations. In the case, which involved a fairly extreme fact pattern, the SDT struck off the offending lawyer, Mr Kumar. 

Mr Kumar, a registered foreign lawyer, was originally before the SDT following a criminal conviction for knowingly employing a person disqualified from work by reason of immigration status. In defending allegations made against him by the SRA, he filed an Answer citing authorities that were variously non-existent, incorrectly cited, or relied upon for propositions they did not support. The SRA identified the errors and expressly directed Mr Kumar to the Divisional Court's landmark guidance in Ayinde v Haringey LBC [2025] EWHC 1383 (Admin), which addressed AI-generated false citations placed before the court and which we considered in a previous blog. Mr Kumar’s response, which purported to correct the errors, was itself drafted using generative AI and contained further false material. He later accepted this was the case and confirmed he was unable to provide copies of the authorities on which he relied.

In light of its findings on both the criminal conviction and Mr Kumar’s use of AI, the SDT struck off Mr Kumar, confirming that it would have imposed the same sanction for his failures arising from the use of AI independently of the underlying conviction. 

Several aspects of the reasoning are of broader significance:

• The SDT stressed that the issue was about a lawyer’s duties in litigation and to the court, which include acting with integrity, making only assertions which are properly arguable and not misleading the court. It was “not necessary” to its conclusion that the erroneous material was generated by or with AI.
• Mr Kumar’s characterisation of his conduct as “technological negligence” and assertion that he “mistakenly believed” AI to be a “lawful drafting aid” were rejected as reflecting a “persistent lack of insight into his misconduct”. It was no answer that Mr Kumar claimed to be unaware of recent regulatory warnings or procedural directives regarding the use of AI in the legal profession. In reaching its decision, the SDT gave very substantial weight to the repeated nature of the misconduct, in particular after errors had been pointed out and explained to him.

The clear message is that AI-generated errors will not be treated as a novel category deserving special leniency, but rather that existing professional standards will be applied. As the SRA Warning Notice and the SDT have confirmed, a lawyer's duty to verify their output remains regardless of whether an AI tool has been used in its production.