AI Hallucination Cases database

Paper · Source
AI at Work

Source: Damien Charlotin · 2026-10-09

The court identified multiple hallucinated citations in Tapscott’s final self-represented appellate brief. It ordered him to provide the cases relied upon or explain the citations, but he did not respond. In combination with his repeated failures to comply with appellate briefing rules and prior warnings, the court dismissed the appeal.

The Court of Appeals identified multiple citations in Grobler’s pro se brief as apparently AI-hallucinated, including three nonexistent Kentucky cases: Hunt v. Smith, 670 S.W.2d 248 (Ky. App. 1984), Bargo v. Bargo, 616 S.W.3d 389 (Ky. App. 2020), and Roberts v. Hensley, 422 S.W.3d 727 (Ky. App. 2013). The court also stated that several citations to real cases did not support the propositions for which they were cited. Treating these errors as a substantive violation of the appellate rules, the court limited review to manifest injustice and affirmed the order. The court separately rejected Grobler’s purported IRS Form 1099-C as plainly unauthenticated and not issued by the IRS, but did not identify that document as an AI hallucination.

Defense counsel admitted using a Westlaw AI tool to identify cases and copying propositions without verifying the authorities or citations. The court found one fictitious case, several false quotations and mis-citations, and at least one materially misrepresented precedent. It held that citing the fictitious case violated Rule 11, but affirmed the magistrate judge's decision not to issue an order to show cause or impose sanctions because Rule 11 sanctions are discretionary, the errors caused no sufficiently demonstrated burden requiring further proceedings, and an admonition was within the court's discretion.

The appellant’s skeleton argument cited two apparent Upper Tribunal decisions—AA (Afghanistan) CG [2024] UKUT 00123 (IAC) and AS (Afghanistan: risk categories) CG [2023] UKUT 00456 (IAC). The judge stated that these appeared to be AI hallucinations and therefore focused on the Respondent’s current policy documents as the relevant country-information sources. The appeal was dismissed, with no separate professional sanction or monetary penalty imposed for the citations.

Plaintiff sought sanctions against Attorney Fred Charles for citing “Mpala v. Segarra, 715 F. App’x 84” instead of the genuine “Mpala v. Segarra, 718 F. App’x 84,” and “LinkCo, Inc. v. Naoyuki Akikusa, 357 F. App’x 180” instead of “367 F. App’x 180.” The incorrect reporter numbers corresponded to unrelated cases, Johnson v. Turnbill and United States v. Philley. The court accepted Charles’s explanation that these were inadvertent typographical errors, noted that the genuine cases were readily locatable, and found no evidence of bad faith, deception, or frivolous legal argument. It denied sanctions under the court’s inherent authority, 28 U.S.C. § 1927, and Rule 11.

Lines of inquiry this paper opens 12

Research framings built by reading the notes related to this paper — the questions it feeds into.

What are the real-world consequences of AI citation hallucinations? How do hallucinated citations emerge in AI scholarly output? What governance mechanisms can effectively constrain widely deployed AI systems? Can AI systems perform peer review as effectively as humans? What gaps exist between benchmark performance and real deployment outcomes? Why do language models hallucinate and how can we prevent it?