AI Chats May Be Discoverable, Even Those Regarding Legal Strategy

Highlights

• AI prompts and outputs, even when created in anticipation of litigation, may be discoverable.
• Courts may look to the privacy policies of the AI tool used to determine the applicability of the attorney-client privilege.
• If the use of AI is not at the direction of counsel, and does not reflect current litigation strategy, its output may not be afforded protection under the work product doctrine.

As some have argued that the use of artificial intelligence technologies shall ultimately become ubiquitous in the practice of law, questions regarding the discoverability of prompts and outputs have been omnipresent in the minds of litigators and sophisticated litigants. A U.S. court has now had the opportunity to address this issue. In United States v. Heppner — a securities and wire fraud case in the Southern District of New York (1:25-cr-503), the court stated, "[o]nly three years after its release, one prominent AI platform is being used by more than 800 million people worldwide every week. Yet the implications of AI for the law are only beginning to be explored."

Heppner's AI Use

In Heppner, the defendant himself allegedly used an AI tool to help him prepare legal memoranda that he then shared with his attorneys. After receiving a grand jury subpoena, Heppner used an AI platform to draft approximately 30 memoranda to his attorneys that detailed proposed defense strategies.

When executing a search warrant, the government seized documents and electronic devices from Hepner's residence. Heppner's counsel informed the government that, among the materials seized, were the AI generated memoranda and asserted privilege over these documents. The prosecution moved to set aside these privilege designations and access the memoranda. In opposition to this motion, Heppner's attorneys argued that this use of AI should be covered by attorney-client privilege or the work product doctrine arguing:

• Attorney-Client Privilege: (1) Heppner used information he received from counsel in his prompts; (2) the memoranda were created to assist Heppner in seeking legal advice from his counsel; and (3) the memoranda were subsequently shared with counsel, making them attorney-client communications.
• Work-Product Doctrine: The AI-generated memoranda were prepared in anticipation of litigation as Heppner created them only after receiving a grand jury subpoena making clear he was the target of the investigation.

Attorney-Client Privilege

The Court held that this privilege only applies to communications between a client and attorney that are intended to be, and are in fact, kept confidential for purposes of obtaining or providing legal advice. The court found that at least two, if not all three, elements were not met in these circumstances:

1. According to this Court, the use of AI was not a communication with Heppner's attorneys; they were communications with the AI tool.
2. The AI tool's written privacy policy, which Heppner agreed to when using the tool, specifically stated that users prompts and the tool's outputs are used to train the platform, and that the developer reserves the right to disclose such data to third parties and governmental regulatory authorities. The Court raised this point at oral argument, finding that there was no expectation that the AI inputs or outputs would be kept confidential.
3. The Court found that it was a "closer call" whether Heppner's use of the AI tool was for the purpose of obtaining legal advice. The use of AI can be equated to communicating with a "highly trained professional who may act as a lawyer's agent within the protection of the attorney-client privilege." However, since Heppner unilaterally used the AI tool, not at the direction of his attorneys, the inquiry was whether Heppner intended to obtain legal advice from the tool, not whether he intended to use the tool's outputs to obtain legal advice from his attorneys.

Finding that at least two of the three factors were not met, the court found that the AI memoranda were not covered by attorney-client privilege.

Work Product Doctrine

In the Second Circuit, the focus of the work product doctrine is to shield the mental processes of the attorney from disclosure. The Court held that the AI memoranda at issue in this case were not prepared at the behest of counsel, thus the qualified protection afforded to documents "prepared by or at the behest of counsel." Heppner's unilateral decision to interact with the AI tool to create these memoranda did not fall into the definition of materials protected by this doctrine.

Moreover, the memoranda did not reflect the metal processes of counsel. While the contents of the memoranda may affect strategy prospectively, they did not reflect the strategy that counsel had at the time they were created. The court found that the work product doctrine is meant to protect current strategies from disclosure, not potential future ones.

Vigilance is Warranted

As the use of AI continues to expand, attorneys must remain up to date on how such material is treated by courts. Knowing that any AI prompts or outputs may be discoverable could affect how lawyers counsel their clients to interact with these tools, particularly once litigation is known or threatened.

As the Heppner court noted, "the implications of AI for the law are only beginning to be explored." As that exploration continues, it will be important for attorneys and AI users to understand the implications.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.