For the past few years, lawyers have repeatedly been warned about the danger of AI “hallucinations,” which most famously include nonexistent cases generated by artificial intelligence and then cited in court filings.
A recent decision from the District of Connecticut makes clear that avoiding fake cases is no longer enough. In Barteca Holdings LLC v. Tacobarn Newtown LLC, No. 3:26-cv-00250 (D. Conn. Aug. 4, 2026), the attorney did something many lawyers might think constitutes adequate AI verification. After writing the brief himself, then using Open Law, Claude and ChatGPT for suggestions, he cross-checked the citations to determine whether the cases existed and Shepardized every case through Lexis.
But he stopped short of verifying whether the cases actually said what the brief represented they said.
The court found problems including incorrect pinpoint citations, quotations that did not appear in the cited cases, inaccurate parentheticals, and legal propositions that were not supported by the authorities cited. The lawyer acknowledged that his verification process had confirmed the existence of the cases, but not the accuracy of the quotations, pin cites, parentheticals, and propositions attributed to them.
As the court put it, quoting Benjamin v. Costco Wholesale Corp., Rule 11 requires lawyers, “at a minimum, [to] read the cases they cite to the Court.” The court concluded that ordinary cite-checking would have revealed the errors.
The result was not merely an admonition. The court imposed a $3,500 sanction and referred the matter to the court’s Grievance Committee. Significantly, the court recognized that this was not one of the more notorious AI cases involving completely fabricated authorities. The cases existed. The attorney also promptly acknowledged his errors and accepted responsibility. Nevertheless, the court concluded that the failure to verify the authorities warranted a meaningful sanction.
The takeaway for lawyers is important: If your AI protocol consists of checking that every cited case actually exists, it is time to revise the protocol. Courts are making clear that AI verification requires the same thing competent legal research has always required: read the case.
Check the quotation. Check the holding. Check the parenthetical. Check the pinpoint citation. And make sure the proposition for which you are citing the case is actually supported by the case.
The question is no longer simply, “Did AI make this case up?”
The question is: “Have I personally verified that this authority says what I am telling the court it says?”
And Barteca makes clear that courts may expect the answer to be yes.