Hernandez v. Kunes

What sanction did the court impose?

Prospective Certificate-of-Use regime imposed. No sanction in this order. Failure to comply with the disclosure requirement may result in sanctions in any future filing. The underlying habeas corpus matter proceeds on its merits.

Why does Hernandez v. Kunes matter for law firms using AI?

Hernandez v. Kunes warrants a careful editorial note before discussing the substance. The Ropes & Gray AI court order tracker, the most-cited public list of judge-by-judge AI orders, attributes this November 5, 2025 Schwab Certificate-of-Use order to Manivannan v. County of Centre, No. 4:21-cv-01359. That attribution is incorrect. The Manivannan docket has had no entries since November 2023, after Centre County and Penn State were terminated as defendants, and contains no AI order. The Schwab order at issue was actually issued in Hernandez v. Kunes, a habeas corpus matter referred to her chambers, on the same November 5, 2025 date. Magistrate Judge Schwab subsequently issued an identical order in Florian v. O’Neill (1:25-cv-02253) on January 29, 2026, confirming that the Hernandez order is part of a chambers template rather than a one-off response to a specific filing in Manivannan.

Substantively, the order is a prospective Certificate-of-Use regime: any party using generative AI in document preparation must file a certificate identifying the specific tool used (the order names ChatGPT and Gemini as examples), the portions of the filing prepared by the AI tool, and a person’s certification that the AI-prepared portions, including citations and legal authority, have been verified for accuracy. The order references the joint formal opinion of the Pennsylvania Bar Association and Philadelphia Bar Association on AI ethics, which is unusual: a federal magistrate operationalizing a state bar’s advisory opinion as a procedural floor in her chambers. That cross-reference is one of the clearest examples of the bar opinion having reach despite being non-binding.

For PA practitioners, the operational implication is that chambers practices vary enough in M.D. Pa. that AI-assisted work cannot be governed by a single firm-wide template. Mehalchick, Caraballo, Latella, and Schwab each issue different disclosure regimes; Schwab’s Hernandez template is the most prescriptive of them on the certification language. A firm-wide AI policy that names the specific tool used and describes the verification step performed by a person, not just that AI was used, will satisfy all four regimes. A policy that only flags AI use generically will not satisfy Schwab’s.