Uploading Litigation Documents to Public AI Tools May Defeat Sealing

Fry v. Fry, No. 26-cv-3469-KSM (E.D. Pa. Aug. 27, 2026)

A new decision from the Eastern District of Pennsylvania adds a distinct wrinkle to the fast-developing law on generative AI in litigation. A prior wave of rulings asked whether AI prompts and outputs are privileged. Fry asks a different question: can you keep a document under seal after you have fed it to a public AI tool? The court's answer was no.

What happened

A pro se litigant moved to seal unredacted materials. Because an earlier order required an AI-use certification, he disclosed that he had uploaded redacted documents to non-enterprise versions of Claude and Claude CoWork. The court seized on that disclosure. It reasoned that material uploaded to Claude is processed on the vendor's servers, and that the vendor's own privacy and training pages make clear the information is not private. The court therefore held that "it would be inappropriate for the Court to seal a document that has already been provided to a private, for-profit research and artificial intelligence company," and denied wholesale sealing.

Why it matters

This is not a privilege holding; the court did not decide waiver, discoverability, or work product. But the practical lesson lands hard: putting sensitive material into a consumer AI tool can forfeit confidentiality protections you later need, including sealing. Interestingly, the court also required the litigant to file a sworn certification identifying the AI tool he used, every prompt he entered, and every document he uploaded—a disclosure process other courts may begin to require.  

Takeaways

1. Route sensitive work through vetted enterprise systems, never public tools.
2. Treat any upload to a consumer platform as a potential loss of confidentiality.
3. Expect courts to require AI-use certifications that surface exactly what was uploaded.