US v. Heppner (S.D.N.Y. 2026) no attorney-client privilege for AI chats [pdf]
A recent U.S. federal court ruling in United States v. Heppner held that chats with AI tools like Claude are not protected by attorney–client privilege or the work product doctrine, because the AI is not a lawyer and its terms of service explicitly disclaim confidentiality. Commenters explore how this reasoning could extend to other online tools such as Google Docs, email, and cloud-based AI assistants, raising concerns that routine use of these services might inadvertently waive legal protections. The debate centers on whether the law should adapt to treat AI-assisted note-taking and research more like traditional privileged preparation, and what technical or contractual safeguards (e.g., self-hosted models or enterprise-grade privacy terms) might be needed to preserve confidentiality.
Case and Ruling Basics
- Court held that chats with Claude in this case are not protected by attorney–client privilege or work-product doctrine.
- Key reasons discussed:
- No attorney involved in the AI chats.
- AI is not a lawyer and explicitly disclaims giving legal advice.
- Terms of service state there is no confidentiality; data may be disclosed and used for training.
- Documents were created before being shared with counsel, so could not be “retroactively” cloaked in privilege.
- Sharing privileged attorney advice with the AI may itself waive privilege over the original communications.
Attorney–Client Privilege vs AI Tools
- One side: AI chats are like talking to a non-lawyer friend; no privilege, by definition.
- Other side: for many users, AI is effectively used as legal research or note-taking, and privilege should focus on case preparation, not the formal status of the “listener.”
- Dispute over whether privilege should evolve to treat certain AI tools as part of the privileged workflow.
Comparisons to Email, Docs, and Phone Calls
- Many note tension between this ruling and everyday use of Gmail/Outlook/Google Docs/Office 365, whose TOS also allow some data access.
- Some argue: using cloud tools doesn’t normally destroy privilege; users reasonably expect confidentiality (similar to phone carriers as “common carriers”).
- Others point out: this judge leaned heavily on the AI provider’s explicit “no confidentiality” language, which could equally threaten privilege for cloud-integrated tools with AI features.
Work Product and Client Notes
- Debate over whether client-created materials for eventual use by counsel should be protected.
- One view: only materials prepared by or at direction of counsel qualify; this decision narrows work-product and overrules a more expansive prior case.
- Another view: strategy notes and research done by an accused in anticipation of litigation should be protected, even without explicit attorney direction.
Implications for Lawyers and Pro Se Litigants
- Concern that pro se defendants effectively lose confidential preparation if AI use is discoverable, while lawyers may still gain some work-product protection.
- Worry that ubiquitous AI integrations (Copilot, Gemini sidebars, etc.) could silently undermine privilege for both lawyers and clients.
Privacy Policies, Logging, and Alternatives
- Heavy emphasis on AI privacy policies: training use and disclosure to authorities undermine any “reasonable expectation of confidentiality.”
- Discussion of “no-log” or self-hosted models, VPN use, and even darknet-style AI services; others note legal compulsion can force logging or disclosure anyway.
- Open question whether a HIPAA-like regime or enclave-based “legal AI” service could eventually satisfy courts’ confidentiality expectations.