Southern District of New York Rules that AI-Generated Documents Are Not Protected By Attorney-Client Privilege

With the recent emergence of generative artificial intelligence (AI), some clients may be asking themselves, ‘Why should I ask a lawyer for legal advice when I can simply ask AI?’ While it is true that AI can be a helpful tool in select situations, AI should not be used in lieu of a lawyer due to its inability to properly support its legal analysis and tendency to use non-existent case law.

In February, a ruling from the federal court in the Southern District of New York confirmed that AI is not a lawyer or a member of a client’s legal team for purposes of attorney-client privilege and the work-product doctrine. This decision is significant because it opens the door for litigants to obtain information about an adverse party’s case by requesting the party’s AI input/output records during discovery.

Below we discuss this decision in United States v. Heppner and provide guidance on minimizing risks and preserving privilege.

United States v. Heppner
In United States v. Heppner, the federal government initiated a federal securities-fraud case against the defendant, Bradley Heppner. Heppner then used Anthropic’s Claude, a third-party generative AI platform—like ChatGPT, Gemini, Perplexity, and Grok—to assess his legal exposure. Specifically, Heppner would input prompts into Claude about the federal government’s investigation that included facts and details he learned from his lawyer, and Claude would generate written responses that Heppner would use to prepare for his discussions with his lawyer. Heppner claimed that he shared Claude’s written responses with his lawyer sometime after their discussions, but it was undisputed that Claude’s written responses were prepared by Heppner on his own initiative—not at his lawyer’s direction.

As a result, the federal government claimed that the nearly 31 AI-generated documents consisting of Heppner’s prompts and Claude’s responses were neither protected by the attorney-client privilege nor the work-product doctrine, and therefore, the AI-generated documents could be obtained by the federal government during discovery and used against Heppner at trial.

The court agreed with the federal government that Heppner’s input and output data from Claude was not protected by the attorney-client privilege for three reasons:

• First, the court held that the AI-generated material was not communications between Heppner and his lawyer or his lawyer’s agent because Claude is a third-party AI platform, not an attorney.
• Second, the court held that the AI-generated material did not memorialize confidential communications because Claude’s privacy policy allows Claude to collect its users’ inputs and Claude’s responses, use of that data for training, and to disclose that data to third parties.
• Third, the court held that the AI-generated material was not generated for the purpose of Heppner obtaining legal advice because Claude clearly states that it “can’t give legal advice.”

As such, Heppner could not withhold the 31-AI generated documents from discovery on the grounds that the documents were privileged attorney-client communications.

Next, the court held that Heppner’s input and output data from Claude were not protected by the work-product doctrine. In doing so, the court explained that the purpose of the work-product doctrine is to protect a lawyer’s strategy. Therefore, any materials that are not prepared by or at the direction of counsel in anticipation of litigation, like Heppner’s input and output data from Claude, cannot be protected from discovery by the work-product doctrine.

How to Minimize Your Risk and Protect Your Privilege During Discovery
In light of the court’s decisions in United States v. Heppner, everyone—not just litigants—should be wary of the information they share with AI generative platforms and should proceed with caution to avoid unintentionally breaching any duty of confidentiality imposed on them by law, contract, or fiduciary relationship.

Companies should also review the terms and conditions of use and privacy policies of AI platforms they elect to authorize for use in their company, as well as their corporate policies on handling sensitive, confidential, and private information, and if necessary, adjust those policies to prohibit the use of that information with AI-generative platforms.

As for current and potential litigants, privilege may be preserved and any risks may be minimized by:

• Prohibiting all use of AI-generative platforms that have policies allowing the retention and disclosure of its users’ data like certain versions of Claude, ChatGPT, and Gemini.
• Avoiding use of AI-generative platforms with inadequate privacy policies to obtain legal advice if litigation may occur or has started.
• Not inputting questions into AI-generative platforms that you are considering asking a lawyer.
• Not giving AI-generative platforms information learned from your lawyer or documents given to you by your lawyer.
• Treating documents generated by AI-generative platforms, including prompts and outputs, as potentially discoverable records that should be managed and preserved appropriately.
• Maintaining a privilege log of AI-generated materials that were created at the direction of your lawyer, which identifies when those materials were created and the basis for the material’s confidentiality, such as a contractual term or fiduciary relationship.

Conclusion
When used appropriately and responsibly, AI is a helpful tool for lawyers, their clients, and pro se litigants to use during and in anticipation of litigation. But AI is not a lawyer, and communications with AI generative platforms are not confidential.

So, the next time you ask yourself, ‘Why should I ask a lawyer for legal advice when I can simply ask AI?’ also ask yourself, ‘Am I fine with this information being used during litigation if and when a lawsuit is filed?’ If the answer to that question is no, then you should forgo seeking legal advice from ChatGPT and consider calling a lawyer.

Please contact Abigail W. White or any member of the Phelps Artificial Intelligence team if you have questions or need advice or guidance.