Attorney AI use
TOV Realty, LLC v. Suarez / Kosel Equity, LLC v. MacGregor
ChatGPT corrupted previously verified citations during a stylistic edit, and the firm’s reviewers did not check them again.
Lawyer and firm sanctioned01
Case background
Material facts
Counsel researched the appeals on LexisNexis, verified the authorities with Shepard’s, and then pasted the drafts into ChatGPT for stylistic editing. ChatGPT changed or added about seven citations. Because the legal propositions remained correct, neither counsel nor the firm’s later reviewers checked the citations again before filing. Amici identified the errors, and counsel then filed corrected briefs. At the show-cause hearing, counsel made “a personal statement, taking full responsibility for the erroneous information, expressing a sincere appreciation for the seriousness of his conduct, and promising that he will be vigilant moving forward.” 355 Conn. at 904.
02
The decision
The court’s ruling
In its first case involving hallucinated citations produced through generative AI, the Connecticut Supreme Court found a competence violation. It required six additional CLE hours, including three on AI, and directed the lawyer and the firm each to donate $1,000 to the Connecticut Bar Institute. It also required a compliance report and provided a courtesy copy of the order to the Statewide Grievance Committee, while specifying that the copy was not a referral for further disciplinary action.
Why it matters
What the decision means
Verification must follow every AI-assisted edit. A source check performed before an AI edit does not validate citations in the final document. That remains true when the AI is asked only to improve organization or style and the underlying legal propositions remain correct.
Primary reading