Doe v. The Trustees of Columbia University in the City of New York
Supreme Court of the State of New York, New York County · N.Y. Sup. Ct. · New York bar guidance
Conduct
Petitioners' counsel filed papers the court says contained AI-generated hallucinated citations; after the reply papers flagged them, counsel withdrew them by letter and apologized.
Consequence
The court declined to sanction counsel, finding it not necessary here, and issued a general warning about AI use; the petition was dismissed on unrelated statute-of-limitations grounds.
Lesson
Here counsel withdrew the citations by letter, apologized again at argument, and described remedial steps; the court then found sanctions not necessary.
Verified September 30, 2026
- Citation
- Doe v. Trustees of Columbia Univ. in the City of N.Y., Index No. 166690/2025, 2026 NY Slip Op 31902(U) (N.Y. Sup. Ct., N.Y. Cnty. May 5, 2026) (Frank, J.)
- Decided
- May 5, 2026
Summary
Five petitioners proceeding as "J. Doe" brought an Article 78 proceeding against The Trustees of Columbia University in the City of New York, Barnard College, and Columbia University, challenging actions the respondents took in response to the petitioners' participation in a May 2025 student protest. Justice Lyle E. Frank's decision addresses, as a postscript, "certain AI-generated hallucinations in Petitioners' papers." According to the decision, after the "fake citations were pointed out in the reply papers," petitioners' counsel wrote a letter to the court acknowledging the false citations, withdrawing them, and apologizing. At oral argument, according to the decision, counsel reiterated the apology, acknowledged that the use of AI without proper checking was wrong, and identified specific steps he and his firm were taking to prevent future problems. The decision does not identify the citations, say how many there were, or name counsel.
- AI tool:
- Unidentified AI (the court refers to "AI-generated hallucinations" in the petitioners' papers, and counsel acknowledged at oral argument that using AI without proper checking was wrong; no tool is named)
How did the court address the AI-related conduct?
The court granted the respondents' motions to dismiss (motion sequences 002 and 003), denied the notice of petition (motion sequence 004), and dismissed the petition. The dismissal rests on the statute of limitations: the petition filed on the last day of the limitations period was verified only by counsel, the petitioners' affidavits came after the period ran, and the court held that without a timely verified petition it lacked jurisdiction. On the citations, the court stated that it "does not deem it necessary to issue sanctions here," and issued what it called "a warning about the use of AI in preparing court documents," stating that it is incumbent on every party, and especially on attorneys, to ensure that AI output is verified and accurate, and that submitting hallucinated or misstated material to courts "is an ethical violation." No sanction was imposed.
Why does Doe v. The Trustees of Columbia University in the City of New York matter for law firms using AI?
The AI discussion in Doe v. Trustees of Columbia University is a postscript to a decision that turned on a procedural defect unrelated to the citations. The petition, filed on the last day of the limitations period, was verified only by counsel; the petitioners’ own affidavits came after the period ran. Justice Frank held that without a timely verified petition the court lacked jurisdiction, and dismissed. The court then turned, under the heading “A Caution on AI Hallucinations,” to citations in the petitioners’ papers that had been “pointed out in the reply papers.”
What the decision records is the sequence of counsel’s response. Counsel wrote to the court acknowledging the false citations, withdrew them, and apologized; at oral argument he repeated the apology, acknowledged (in the decision’s words) that the use of AI without proper checking was wrong, and “identified the specific ways that steps were being taken by him and his firm to prevent future problems.” The court then stated that it “does not deem it necessary to issue sanctions here.” The decision does not say which of these steps, if any, was decisive.
The warning the court issued is framed generally rather than at this lawyer: “it is incumbent on every party communicating with a court, but especially incumbent on attorneys, to ensure that any output generated by AI is verified and accurate,” and “[t]he submission of hallucinated or misstated material to courts is an ethical violation.” Firms documenting compliance may wish to consider how this court treated a prompt written withdrawal and a description of remedial steps, while noting that the decision is a single trial-court ruling that the Law Reporting Bureau marks as not selected for official publication.
Implications for your firm
Operational steps a firm reading this case may wish to consider documenting. Strategic and rule-application calls belong to your firm's attorneys.
- Document the correction steps taken once a citation problem is flagged. In this case counsel's letter withdrawing the citations, the apology at oral argument, and the specific remedial steps he described preceded the court's decision not to sanction.
- Review AI-assisted drafts before filing. The court stated that it is incumbent on attorneys to ensure any AI output is verified and accurate, and described submitting hallucinated or misstated material as an ethical violation.
- Consider that here the citations were pointed out in the reply papers, before the court addressed them.
Sources
Primary sources
- The decision does not identify which filing contained the hallucinated citations, what the citations were, or how many there were; counsel's letter to the court and the reply papers that flagged the citations were not read.