Hallucinating the Facts: Court Sanctions Counsel for Fabricated Deposition Quotes
Pauliah v. Univ. of Mississippi Med. Center, et al., (S.D. Miss. Dec. 30, 2025)
Why This Alert Is Important
While previous high-profile AI sanctions focused on "hallucinated" case law (fictitious judicial opinions), this ruling highlights another dangerous risk: the fabrication of facts. For legal professionals, this serves as a critical warning that delegating document summarization to generative AI without line-by-line verification can lead to severe reputational damage and monetary sanctions.
Overview
In Pauliah v. University of Mississippi Medical Center, Plaintiff filed a declaration in opposition to a motion for summary judgment that appeared to contain smoking-gun evidence. The 40-page, single-spaced document was peppered with what appeared to be direct, verbatim quotations from deposition transcripts of the defendant's employees, complete with specific page and line citations.
When defense counsel attempted to locate the quoted passages in the actual transcripts, they found that the statements simply did not exist. The court subsequently held a Rule 56(h) hearing to determine whether the declaration was submitted in bad faith. During the hearing, a blame game ensued: the plaintiff’s attorney claimed the client drafted much of the document, while the client, after initially denying it, admitted to using generative AI to summarize the depositions.
Neither the client nor the attorney had verified the AI-generated summaries against the actual record before signing and filing the declaration. The court noted that this was a novel and "unacceptable" act, as AI had "hallucinate[d] the facts" rather than the law.
Ruling Summary
- “Grossly Mischaracterized” Testimony
The court found that submitting "grossly mischaracterized" testimony and "manufactured citations" was a direct threat to the integrity of the judicial process. The ruling noted that “manufacturing ‘facts,’ then presenting them to the Court as genuine, threatens to corrupt the Court's analysis and undermine the integrity of the judicial process at the summary judgment stage.” - Non-Delegable Duty to Verify AI Output
The ruling emphasized that while AI may be a "terrific capability" for document summarization, it does not absolve counsel of their ethical obligations. The plaintiff’s attorney admitted that he "had a responsibility to review whatever [Plaintiff] prepared,” and therefore his failure to detect the “completely fabricated deposition testimony [was] hard to fathom, as [he] not only attended the depositions, he took them, and he obtained copies of the transcripts.” - Imposition of Monetary and Educational Sanctions
Finding the conduct of both the plaintiff and his counsel to be "unacceptable," the court imposed a tiered sanction structure. The attorney was ordered to pay $4,000 to opposing counsel and attend a three-hour CLE course specifically focused on "hallucinatory citations generated by AI." Additionally, the plaintiff was personally sanctioned $1,000, reinforcing that clients who use AI to generate evidence are also subject to the court’s power to penalize bad-faith filings.
Pauliah is an eyebrow-raising ruling for professionals navigating generative AI in evidentiary workflows. What began as an authoritative declaration with pinpoint quotations and citations unraveled into a cautionary example of unchecked AI summarization, ending in a blame-shifting dispute between plaintiff and counsel.
Risks will intensify as AI agents increasingly rely on upstream AI outputs, compounding hallucinations. For legal and eDiscovery practitioners, the lesson is clear: In order to maintain defensibility, AI-assisted review and summarization must meet the same standards as any discovery process, including transparency into when and by whom AI is used, auditable workflows (similar to chain-of-custody), and validation of facts, quotes, and citations.
Ultimately, Pauliah is a stark reminder that while AI can accelerate analysis and drive efficiency, it cannot assume professional responsibility. Bree Murphy, CEDS,Co-Founder, eDiscovery Chicks, and Strategic Account Executive, Exterro
Data Privacy Tip
This ruling completes the warning arc for the legal industry: AI is not a substitute for the manual red-line verification of transcripts and case law. When using AI for summarization, legal teams should treat the output as a draft that requires strict corroboration against "ground truth" source documents.
This alert is for informational purposes only and is not legal advice.
