The Brief
Issue 15 · 28 September 2026
He used AI to fix his AI mistake. That went wrong too.
The SDT's first AI strike-off used AI to explain itself, and got that wrong too. Ireland stands up a dedicated AI regulator. A ChatGPT summary nearly derails a murder appeal in New Mexico. And Australia opens a formal review of whether its laws are ready.
The Solicitors Disciplinary Tribunal handed down its first ruling addressing a lawyer's use of generative AI this month, in the case of registered foreign lawyer Abhishek Kumar. Kumar was already facing separate SRA disciplinary charges, over a 2024 criminal conviction for employing a worker disqualified by his immigration status. Asked to answer those charges in writing, he used AI to draft his response. That response contained misleading quotations and case citations that do not exist.
Once the regulator flagged the fabrications, Kumar used AI again, to draft his explanation of the first mistake. The explanation itself contained further fabricated material. The Tribunal found his culpability for the AI errors “very high”, holding that he had “embarked on the use of AI without giving any proper thought to the consequences.” It struck him off the register, stating plainly that the sanction would have been the same either way: “the sanction it reached was the same sanction it would have imposed if either allegation had appeared before the tribunal without the other.”
This was, in the Tribunal's own words, “the first time that a lawyer's use of artificial intelligence or large language models in legal proceedings had been litigated before the Tribunal.” What made it worse than a first offence is that Kumar's attempt to account for his first AI mistake, using the same tool the same way, produced a second one. Checking the tool's output once was not the failure. Never checking it, twice in a row, was.
Ireland stands up its AI regulator
The Regulation of Artificial Intelligence Act 2026 became law on 21 July, signed by President Connolly. It establishes the AI Office of Ireland as an independent statutory body and the country's central coordinating authority under the EU AI Act.
Paul Byrne, previously of the Medical Council of Ireland, was confirmed as the Office's first Chief Executive Officer. It became operational on 2 August, coordinating Ireland's existing sectoral regulators rather than replacing them, matching the AI Act's distributed-authority model.
Ireland now has a single, named point of contact for AI Act compliance questions, rather than a patchwork of sectoral regulators each guessing at their own remit. For any practice advising clients in or through Ireland, that is one fewer place a compliance question can fall between two desks.
A ChatGPT summary derails a murder appeal
The New Mexico Supreme Court held Santa Fe attorney Stephen Aarons in contempt and fined him $5,000 after his appellate brief for a murder conviction cited testimony from witnesses who do not exist. Aarons had fed trial materials into ChatGPT expecting a summary, and filed what it returned without checking it.
Justice C. Shannon Bacon's response from the bench captured how little patience is left for this excuse: “Counsel, do you watch the news? Do you listen to the radio? Do you read anything about what's going on in the world? Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day.”
The court struck his filings, referred him to the state's attorney disciplinary board, and reassigned the underlying appeal, for client Oscar Renee Sandoval, to a public defender.
This is not a discovery motion or a costs order. It is a murder appeal, and the fabrication reached the court on the first filing, not after a second chance. The stakes of not checking AI output scale with what is actually at risk for the client, and here that was a man's conviction standing or falling on witnesses who were never real.
Australia opens a formal review of its AI laws
Australia's Parliament appointed a Joint Select Committee on Artificial Intelligence to examine whether existing law, including copyright, consumer protection, national security and data sovereignty frameworks, is adequate for AI. The 12-member committee closed submissions on 14 September and must report by 30 November.
This sits alongside a broader shift already under way. In July the Prime Minister signalled plans to legislate Australian AI standards rather than rely solely on voluntary guidance, and mandatory disclosure obligations for automated decision-making using personal data take effect under the Privacy Act on 10 December.
None of this is enforceable yet. It is a committee taking submissions, not a finished statute. But the direction is set, and firms already running AI on client or personal data have a two-and-a-half month runway before the first hard date lands.
Our read
Four jurisdictions, four different stages of the same problem. The UK is now enforcing a verification duty tribunal by tribunal. Ireland just stood up the regulator built to coordinate exactly this. A murder appeal in New Mexico shows what happens when nobody checks, whatever the stakes. Australia is still writing the rules the other three are already enforcing. Whichever stage a firm's own jurisdiction is at, the underlying requirement does not change: an answer needs a real, checkable source behind it before anyone relies on it, not after a tribunal or a court finds out it did not have one.
Get the next issue in your inbox