Mullins v. Duquesne University of the Holy Spirit
U.S. District Court, Western District of Pennsylvania · W.D. Pa. · Pennsylvania bar guidance
Conduct
Pro se reply brief contained a phantom Long v. SEPTA quote, an altered quotation, and improper bracketing in a Foman v. Davis cite.
Consequence
Reply stricken with leave to refile; AI disclosure certificate required for any future filing using generative AI.
Lesson
Phantom quotes draw scrutiny in W.D. Pa. even from pro se filers. Voluntary AI disclosure protocols are now the safer default.
Verified May 14, 2026
- Citation
- Mullins v. Duquesne Univ., No. 2:25-cv-01366-MJH (W.D. Pa. Dec. 5, 2025) (Horan, J.)
- Decided
- December 5, 2025
Summary
Pro se plaintiff David Mullins filed a reply brief that contained a phantom quote attributed to the Third Circuit's decision in Long v. SEPTA, an altered quotation that changed "the same" to "identical," and improper bracketing in a citation to Foman v. Davis. Defendants moved to strike. Judge Marilyn J. Horan granted the motion, struck the reply, and ordered that any future filing using generative AI must include a certificate disclosing the specific AI tool used, the portions generated by AI, and certification that all citations and authority were verified against primary sources before filing.
- AI tool:
- Not named in the order striking the brief; plaintiff's subsequent re-filed reply identified ChatGPT and Claude
What sanction did the court impose?
Reply brief stricken with leave to refile by 2025-12-15. Prospective AI disclosure certification ordered for any future filing in the matter using generative AI. No monetary sanction imposed.
Why does Mullins v. Duquesne University of the Holy Spirit matter for law firms using AI?
Mullins is the rare AI hallucination case where the diagnostic was not a fabricated case but a fabricated quotation inside a real case. The cited authorities (Long v. SEPTA, Foman v. Davis) exist; the quoted language does not appear in the originals. That failure mode is harder to catch than a fabricated citation because the cite-checker tool will return a positive match for the case name and the reporter, and only a primary-source read will surface the textual divergence. Judge Horan’s prospective disclosure-certificate requirement is narrow in scope, applying only to future filings in this matter, but it tracks the standing-order pattern emerging across the Pennsylvania federal districts: disclose the tool, identify the AI-generated portions, and certify accuracy against primary sources before filing.
For a small Pennsylvania firm with W.D. Pa. matters, the operational implication is that quoted language deserves a separate verification pass beyond cite-checking. AI-assisted draft generation routinely introduces small textual edits that fall outside what a reporter or a citation-validation tool will flag. Building a “verify quoted language verbatim” step into the brief-finalization checklist is a low-cost protection that addresses Mullins-style errors before they reach the court.
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 a citation-verification step that confirms quoted language against the primary source, not just a citation match.
- Consider requiring an AI-use certification on all reply briefs in W.D. Pa. matters before Mullins-style orders proliferate to other judges.
- Train staff on bracketing and quotation accuracy as a separate skill from cite-checking; AI-assisted drafts often introduce silent textual edits.