Talking to AI About Legal Strategy May Waive Privilege

A recent decision from the U.S. District Court for the Southern District of New York makes clear that communications with consumer generative AI tools are not protected by attorney-client privilege or the work-product doctrine. In United States v. Heppner, the court held that a defendant’s independent use of a public AI platform to analyze legal strategy resulted in no privilege attaching to the resulting materials. When AI is used in connection with legal analysis or strategy, prompts and outputs should be developed only with and at the direction of legal counsel.

In United States v. Heppner, a criminal defendant turned to the consumer version of Anthropic’s Claude to work through aspects of his legal defense without first consulting counsel. When the government argued that doing so waived privilege, Judge Jed Rakoff agreed.

Essentially, information that might have been privileged if communicated directly and confidentially with counsel was deemed discoverable because the defendant first provided it to a consumer-grade AI tool. A straightforward look at the basics of attorney-client privilege explains why.

Attorney-client privilege protects confidential communications between a client and an attorney made for the purpose of obtaining legal advice. No matter how fluent or persuasive AI may sound, it is not a licensed attorney. Because the defendant used the AI tool on his own initiative, without involving counsel at the outset, the communications did not qualify for privilege. That same lack of attorney involvement also defeated any claim of work-product protection, which generally requires materials to be prepared by or at the direction of counsel in anticipation of litigation.

There was also no reasonable expectation of confidentiality. The AI provider’s publicly available terms of use permitted retention and potential disclosure of user data. By inputting sensitive information into a consumer AI platform, the defendant effectively disclosed that information to a third party. Once that disclosure occurred, privilege could not be restored simply by later sharing the AI-generated materials with counsel.

Do

• Assume AI inputs and outputs may be discoverable, especially when using public or consumer tools
• Involve counsel before using AI to analyze legal risk, strategy, or advice
• Limit AI use to non-legal, non-sensitive tasks

Don’t

• Input advice received from counsel or other privileged information into AI tools
• Assume paid or “secure-labeled” AI tools automatically preserve privilege
• Expect privilege to apply retroactively just because AI-generated materials are later shared with counsel

AI is undoubtedly powerful, but it is not a substitute for legal counsel. When legal issues are involved, it pays to be cautious and to involve counsel first.