Federal courts have begun issuing concrete rulings on the discoverability of AI-assisted work and the answers should prompt immediate action inside your organization.

The question is no longer hypothetical. Three federal court rulings in early 2026, all issued within weeks of each other, have established the first real legal framework for how attorney-client privilege and work product protection apply to AI-generated content. The core message is clear: traditional privilege principles apply with full force to AI-generated materials. What matters is how and where AI is used.

These decisions should be required reading for any executive, general counsel, or compliance professional whose team relies on AI tools, which is to say, nearly everyone.

The Landmark Ruling: Public AI Tools Are Dangerous Territory

In United States v. Heppner, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026) (bench ruling Feb. 10, 2026), Judge Rakoff issued what is now the most significant ruling to date on this issue. The case involved a criminal defendant who had used a consumer AI platform without the involvement of counsel. Judge Rakoff's conclusion was unambiguous: doing so destroyed any reasonable expectation of confidentiality. Both attorney-client privilege and work product protection were forfeited.

This ruling reinforces basic principles. Attorney-client privilege protects confidential communications with your lawyer for purposes of legal advice. A public AI system is not your lawyer or your lawyer's agent. And virtually every major consumer AI platform explicitly disclaims confidentiality in its terms of service, often reserving the right to use your inputs to train its models. Similarly, the work product doctrine only protects materials prepared by counsel or at counsel’s direction in anticipation of litigation. Work prepared by the non-attorney client, even when using an enterprise version of an AI platform, that was not done at counsel’s direction, would not be entitled to protection from disclosure either.

Notably, Judge Rakoff left open the possibility that enterprise AI tools, operating under robust confidentiality agreements and used at counsel's specific direction, might qualify for protection under the Kovel doctrine, which extends privilege to certain third-party agents acting under attorney supervision. But that question remains unresolved, and organizations should not assume enterprise tools provide a safe harbor without careful structuring.

The Counterpoint: Work Product Has a Different Standard

Issued the same day as Heppner, a Michigan federal court reached a different result in Warner v. Gilbarco, Inc., 2026 WL 373043 (E.D. Mich. Feb. 10, 2026). There, a pro se litigant used ChatGPT to prepare litigation materials. The defendants argued that uploading work into ChatGPT had waived work product protection. The court disagreed.

The reason these decisions do not conflict is rooted in the different legal standards that apply. Work product protection is waived not by third-party disclosure generally, but by disclosure to an adversary or “in a way likely to get in an adversary’s hand.” The court held that AI tools are instruments, not persons and certainly not adversaries and that a pro se litigant is entitled to assert work product protection. That framing allowed the work product claim to survive.

When Does AI Use Retain Protection? The Three-Factor Test

Drawing on these decisions, courts are focused on three core questions:

1. Confidentiality: Did the AI tool maintain reasonable confidentiality of user inputs, or did its terms of service defeat confidentiality by permitting third-party access or model training?

2. Counsel Direction: Was the AI use directed by counsel, or did a client or non-attorney employee act unilaterally?



3. Litigation Purpose: Do the materials reflect an attorney's mental impressions and were they prepared in anticipation of litigation?

Steps Your Organization Should Take Now

This is not an area where a wait-and-see approach is prudent. The law is moving quickly, and the cost of a privilege waiver, particularly in high-stakes litigation, can be severe. We recommend the following immediate steps:



•         Audit your AI tools. Review AI platforms currently in use. Identify which are consumer-grade (open) versus enterprise-grade (closed) and map them against the contexts in which they are being used.

•         Restrict legal and litigation inputs. Prohibit employees from entering privileged information, litigation strategy, or legal analysis into any non-enterprise AI platform. This policy should be written, communicated, and enforced.

•         Involve counsel before using AI in disputes. If litigation is active or reasonably anticipated, any AI use in connection with that matter should be structured by counsel, not left to individual employees. The Kovel doctrine may offer a path to privilege protection but only if the use is properly directed and documented.

•         Update your policies and contracts. AI governance policies should be updated to specifically address these risks. Provisions that were adequate two years ago almost certainly are not adequate today.

Technology advances on its own schedule. Privilege law does not. We have been closely tracking this area of law and are available to help you assess your organization's exposure, update your internal policies, and structure your AI use to preserve the protections that matter.

We would welcome the opportunity to discuss the implications of these decisions for your business. For any questions, please contact Joshua Leader at (212) 486-2400 | jleader@leaderberkon.com.

© 2026 Leader Berkon Colao & Silverstein LLP | Attorney Advertising | This alert is for informational purposes only and does not constitute legal advice.