Do You Waive Privilege by Using AI? Two Federal Courts Say It Depends

Key Takeaways

• Courts split on AI privilege based on user direction.
• Confidentiality of AI platform dictates privilege status.
• Pro se AI use treated as internal mental impressions.
• Criminal case deemed non‑privileged due to lack of counsel oversight.
• Practitioners must update custodial interviews for AI usage.

Summary

In February 2026 two federal judges tackled whether content generated with publicly available generative AI tools is shielded by attorney‑client privilege or the work‑product doctrine. Magistrate Judge Patti, hearing Warner v. Gilbarco, treated AI as a mere drafting tool and protected the pro se plaintiff’s AI‑assisted notes as work product. In contrast, District Judge Rakoff in United States v. Heppner deemed a criminal defendant’s Claude‑generated strategy reports non‑privileged because the AI’s privacy policy allowed retention and the use was not counsel‑directed. The split underscores that privilege hinges on confidentiality expectations and who controls the AI interaction.

Pulse Analysis

Generative AI has moved from experimental novelty to a routine drafting aid for lawyers and litigants alike. The two February decisions illustrate how courts are applying long‑standing privilege principles to this new environment without rewriting the rules. Judge Patti’s Warner ruling frames AI as a neutral tool, likening it to word processors, and therefore preserves work‑product protection when the user is the litigant. By contrast, Judge Rakoff’s Heppner opinion treats the AI platform as a third‑party repository, emphasizing the privacy policy and the absence of attorney direction as grounds to waive privilege. This doctrinal tension highlights the importance of the confidentiality expectation attached to the AI service.

The divergent outcomes rest on two factual pillars: who initiates the AI query and what the AI provider promises about data retention. In civil contexts, if counsel directs the AI or the platform guarantees confidentiality, privilege is more likely to survive. In criminal matters, where the defendant acts independently and the provider’s terms allow data disclosure, courts are inclined to deem the output non‑privileged. The decisions also raise broader questions about whether AI‑generated prompts, outputs, or integrated drafting assistance qualify as electronically stored information under Rule 26, and how enterprise‑grade AI tools differ from public services.

Practitioners must now embed AI considerations into their discovery protocols. Updating custodian interviews to capture tool usage, reviewing privacy policies, and documenting attorney oversight are essential steps to preserve privilege. Firms should also assess preservation obligations for AI interactions, especially when platforms retain user inputs. As AI becomes embedded in collaboration suites, the line between internal notes and third‑party disclosures will blur, prompting courts to demand concrete factual records. Until appellate guidance emerges, a proactive, fact‑driven approach will shield clients from inadvertent waiver and keep litigation strategies secure.