The Landscape Has Shifted — And Split

In a previous client alert, we reported on United States v. Heppner, No. 25 Cr. 503 (S.D.N.Y. Feb. 17, 2026), in which Judge Jed S. Rakoff of the Southern District of New York ruled that a criminal defendant’s communications with Anthropic’s Claude AI platform were not protected by either the attorney-client privilege or the work product doctrine. That decision reverberated through the legal community, suggesting that any use of a publicly available AI tool to analyze legal issues by a client could destroy privilege protections entirely.

One week earlier, however, a federal court in the Eastern District of Michigan reached the opposite conclusion. In Warner v. Gilbarco, Inc., No. 2:24-cv-12333 (E.D. Mich. Feb. 10, 2026), Magistrate Judge Anthony P. Patti denied a motion to compel production of a plaintiff’s AI-assisted litigation materials, holding that the use of ChatGPT did not waive work product protection. Judge Patti reasoned that “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background,” (emphasis in original) and that work product waiver requires disclosure “to an adversary or in a way likely to get in an adversary’s hand.”

The Key Conflict

The two orders rest on fundamentally different analytical frameworks:

In Heppner, Judge Rakoff found that sharing information with Claude destroyed confidentiality for a number of reasons. While the court discussed the fact that Anthropic’s privacy policy permits data collection, use for training purposes, and disclosure to third parties—including “governmental regulatory authorities”—this was far from the sole reason the court found that neither the attorney-client privilege, nor work product protection, applied to protect the defendant’s use of AI to assist with his legal defense strategy. In addition to reasoning that no “reasonable expectation of confidentiality” could exist when a platform’s terms expressly reserve the right to disclose user inputs, the court further emphasized that all recognized privileges require “a trusting human relationship” with “a licensed professional who owes fiduciary duties and is subject to discipline”—a relationship that cannot exist between a user and an AI platform. (internal citation omitted)

In Warner, Judge Patti took the view that the work product doctrine protects a litigant’s “internal analysis and mental impressions—i.e., her thought process,” and that treating AI use as a waiver “would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed.” The court characterized the defendant’s pursuit of AI-related discovery as “a fishing expedition” and “a distraction from the merits of this case.”

Important Distinctions

While the holdings appear squarely contradictory, several factual distinctions bear noting. In Heppner, the defendant used Claude independently, without direction from counsel, to develop defense strategy—and the court found this critical, noting that had counsel directed the AI use, “Claude might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer’s agent within the protection of the attorney-client privilege.” In Warner, the plaintiff used ChatGPT as a drafting and analytical tool in connection with active litigation, and the court treated this use as analogous to any other software-assisted work product preparation. Additionally, Heppner involved a claim of attorney-client privilege over client-to-AI communications, while Warner focused specifically on the work product doctrine—which historically carries a higher waiver threshold.

What This Means for Our Clients

Until appellate courts resolve these different outcomes, businesses and individuals face genuine uncertainty about whether AI-assisted legal analysis will be protected from discovery. The safest course is to assume the more restrictive Heppner framework could apply in any jurisdiction.

Practical Steps to Protect Privilege

We recommend that clients implement the following measures immediately:

  • Do not use publicly available AI tools to analyze legal questions or develop legal strategy without express direction from counsel. The Heppner court strongly suggested that attorney-directed AI use might receive privilege protection, but independent client use will not. Any legal analysis by a client should be conducted at the direction of counsel or through enterprise AI platforms with appropriate confidentiality safeguards. However, even then, this may not protect the AI-generated output from discovery.
  • Audit and disable AI transcription and recording features during calls with your attorneys. Many conferencing platforms (Zoom, Teams, Google Meet) now include AI-assisted transcription, summarization, and note-taking features that may transmit privileged communications to third-party servers. Under the Heppner rationale, any transmission of privileged content to a third-party platform with broad data-use policies could constitute waiver. Therefore, the safest route is to avoid using these tools entirely during calls where privileged information will be exchanged.
  • Review AI platform terms of service and privacy policies before inputting any sensitive business or legal information. The Heppner court relied heavily on Anthropic’s privacy policy—specifically its provisions allowing data collection, training use, and third-party disclosure—to find that no reasonable expectation of confidentiality existed. Where enterprise agreements with AI platforms provide contractual commitments against using user inputs for AI training and third-party sharing, the privilege calculus may differ.
  • Establish clear internal policies governing employee use of AI in connection with any matter that is the subject of litigation, investigation, or legal consultation. Document that any AI use for legal purposes must only occur at the direction of and under the supervision of counsel, so as to potentially bring such use within the agency framework recognized in Heppner.
  • Maintain AI use within privileged workflows by routing all legal AI queries through counsel’s own systems. If attorneys direct the use of AI tools as part of their own work-product preparation—rather than clients using AI independently—the resulting materials are far more likely to retain both privilege and work product protection under either court’s framework.

Looking Ahead

These conflicting orders underscore that the law governing AI and privilege is in its infancy and demonstrates that courts are still developing frameworks for evaluating AI-assisted legal work. Neither decision is binding beyond its own district, and appellate guidance may take years to materialize. Until then, organizations should assume that at least some courts may treat AI platform disclosures as privilege-waiving communications. In the interim, the prudent approach is to treat AI interactions involving legal and confidential subject matter with the same caution one would apply to any communication with an unretained third-party. We will continue to monitor developments in this rapidly evolving area and provide updates as additional courts weigh in.

For questions about how these rulings may affect your organization’s AI policies, litigation strategy, or internal protocols, please contact any member of Winthrop’s Business & Commercial Litigation team.

June 1, 2026