Judge Allows Bar-Journal AI-Error Article to Avoid Sanctions
Per Reason (Volokh Conspiracy), Chief Judge Martin Reidlinger (W.D.N.C.) issued a November 10, 2025 show-cause order requiring plaintiff's counsel to explain numerous errors in court filings, including hallucinated citations and hallucinated quotations, in Doe v. Univ. of N.C. Sys. At a November 19, 2025 hearing, the counsel admitted the errors, attributed them to misuse of AI tools, and offered to write an explanatory piece for the North Carolina State Bar Journal. Per Reason, the published article - 'Guarding Against AI Errors: Ethical Risks for NC Attorneys' by Fred W. DeVore III and Rob Wilder (N.C. State Bar J., Summer 2026) - fell short of the court's expectations: Chief Judge Reidlinger found it minimized the scope of errors and did not fully acknowledge the impact on opposing counsel and the court. The judge nevertheless declined sanctions, citing the attorneys' good-faith expressions of repentance and their long history of exemplary conduct.
What happened
Per Reason (Volokh Conspiracy, Eugene Volokh), Chief Judge Martin Reidlinger (W.D.N.C.) issued a November 10, 2025 show-cause order after identifying multiple errors in plaintiff's counsel's submissions in Doe v. Univ. of N.C. Sys., including:
- •citing two cases that do not appear to exist (hallucinated citations)
- •quoting material that does not exist in the cases purportedly quoted (hallucinated quotations)
- •mischaracterizing holdings of cited cases
- •citing cases with no bearing on the proposition cited
- •failing to provide pinpoint citations. At a November 19, 2025 hearing, the plaintiff's counsel admitted these errors and attributed them in large part to misuse of AI software and failure to verify AI outputs. The court agreed that a bar journal article explaining those errors could help other lawyers
The bar journal article and the court's response
Fred W. DeVore III and Rob Wilder published "Guarding Against AI Errors: Ethical Risks for NC Attorneys" in the N.C. State Bar Journal (Summer 2026). Chief Judge Reidlinger declined to impose sanctions, finding "their expressions of repentance are made in good faith." However, per Reason's reporting of the opinion, the court expressed clear disappointment: the article "falls short of the Court's expectations," minimized the scope of the errors - five documents with citation errors - and failed to fully acknowledge the impact on opposing counsel and the court. As the court noted, one attorney wrote: "The cases I cited were correct, but the quotes were hallucinated. I checked the cases but not the quotes" - which the court rejected as an incomplete understanding of the verification obligation.
The court also noted the article "unironically concludes with a purported quotation that is neither cited nor attributed to any speaker" - illustrating the very verification failure the article was meant to address.
Editorial analysis - technical context
Observed patterns in similar incidents show that generative models frequently produce plausible but incorrect citations and quotations when used as legal drafting assistants without verification. The court's commentary underscores a critical distinction: checking that a cited case exists is not the same as checking that a case contains the quote or supports the proposition cited.
What to watch
- •Whether other state bars or courts cite this opinion when adjudicating AI-related errors in pleadings or filings.
- •If bar journals and continuing-legal-education providers publish practical verification guidance for attorneys who use generative tools.
- •Whether courts begin treating the quality and completeness of remedial disclosures as a factor in sanctioning determinations.
Scoring Rationale
Notable legal-ethics ruling that surfaces a nuanced judicial standard: remedial disclosures must accurately describe the error and its impact, not minimize it. The irony that the remedial article itself contained an uncited AI-sounding quote sharpens the practical lesson. Relevant to all practitioners using generative tools in professional or regulated contexts.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
