Client Alert: SDNY Rules AI‑Generated Documents Are Not Privileged

The United States District Court for the Southern District of New York has issued a ruling on how courts should treat materials created through public generative AI platforms. In United States v. Heppner, Judge Jed S. Rakoff held that documents a criminal defendant generated using Anthropic’s generative AI tool Claude were not protected by the attorney‑client privilege or the work product doctrine.

This landmark ruling marks a decision of first impression on whether unsupervised client use of public AI tools in connection with an ongoing legal matter can receive traditional privilege protections. The court was unequivocal: “the answer is no.”

Background

Bradley Heppner, an executive charged with securities and wire fraud, created roughly 31 AI‑generated documents using Claude after receiving a grand jury subpoena. These materials outlined potential defense strategies and legal arguments. Although he later shared them with his attorneys and claimed privilege, his counsel acknowledged that Heppner generated the AI documents independently and not at his lawyers’ direction.

The government moved for a ruling that these documents were not privileged—arguing they were neither confidential nor created for the purpose of obtaining legal advice. Judge Rakoff granted the motion from the bench on February 10, 2026 and later issued a written memorandum explaining his reasoning.

Why the Court Found No Attorney‑Client Privilege

Judge Rakoff applied the traditional three‑part test: to be privileged, a communication must (1) be between privileged parties, (2) be intended and maintained as confidential, and (3) be made for the purpose of obtaining legal advice. The AI‑generated materials failed each of these prongs.

1. AI Platforms Are Not Attorneys

The court held that Claude is not an attorney, and Heppner’s interactions with it were equivalent to discussing legal matters with a non‑lawyer third party which does not create a privileged communications.

2. No Reasonable Expectation of Confidentiality

The ruling emphasized that use of a public AI tool under terms allowing provider access defeats any reasonable expectation of confidentiality. Because confidentiality is a cornerstone of privilege, its absence was fatal.

3. Not Created for the Purpose of Obtaining Legal Advice

Heppner produced the AI documents on his own initiative, not at the direction of counsel. As the court stressed, privilege cannot be created “post hoc” by later sending unprivileged materials to attorneys.

Why the Work Product Doctrine Also Failed

The work product doctrine protects documents prepared by or for counsel in anticipation of litigation. Heppner’s AI documents were prepared independently, without guidance from counsel, and thus did not reflect counsel’s mental impressions or litigation strategy. Courts do not extend work‑product protection to layperson‑generated research, whether done via the internet or via AI.

Key Takeaways for Clients

The Heppner ruling sends a clear message: Using a public generative AI tool to analyze or summarize legal matters, without counsel’s direction or confidentiality, will likely cause the search and its results to be subject to discovery and not covered by privilege.

What This Means for Organizations and Individuals

• Public AI tools are third parties. Using open, consumer‑grade AI tools for sensitive legal matters introduces immediate privilege concerns and risks.
• Counsel direction matters. Privilege is far more defensible when counsel explicitly directs and supervises AI‑assisted work.
• Enterprise‑grade AI tools are safer. Privilege analysis may differ when organizations use contracted AI systems with confidentiality obligations and restricted data‑use terms.
• Privilege cannot be retroactively created. Sending AI‑generated summaries or analyses to counsel after the fact will not convert them into privileged documents.
• Careful privilege logging is essential. When AI is used under counsel direction, documentation must clearly show that the work was commissioned for the purpose of providing legal advice.

What Clients Should Do Now

To reduce risk in light of Heppner:

1. Avoid using public AI tools for legal‑adjacent tasks, including drafting factual summaries, timelines, potential defenses, or narratives about ongoing investigations.
2. Consult counsel before using any AI system to generate materials involving legal content.
3. Implement clear policies governing employee use of AI, distinguishing between general productivity tasks and matters involving regulated, privileged, or sensitive information.
4. Adopt enterprise AI systems with strict confidentiality and data‑handling provisions when AI must be used in legal workflows.
5. Train employees – especially executives and those involved in investigations – on the privilege risks highlighted by Heppner.

Conclusion

The SDNY’s decision in United States v. Heppner is a landmark in the developing law of AI and privilege. While it does not hold that generative AI can never be used in privileged work, it underscores that traditional privilege rules apply regardless of technology. Unsupervised use of public AI tools, especially for legal purposes, carries significant risk.

If you have questions about how AI use may impact privilege, or how to structure your organization’s AI workflows safely, we are available to assist.