In a first of its kind ruling, a Judge in the U.S. District for the Southern District of New York held that documents that a defendant (client) created using generative AI and sent to his lawyer were not protected by privilege, nor were they protected by the work product doctrine. Like many rulings, the determination was driven by the facts of the case, so it should not be assumed that the ruling will act as a blanket barrier to using AI and being afforded privilege protection, but the contours of the extent of privilege for use of AI will continue to play out in the months and years to come. For now, it is critical for corporations to understand that employee use of AI and the resulting documents or information generated as a result of such use may not be privileged, and so will be discoverable and potentially usable in litigation against the corporation.

Background

The defendants in the above-referenced case was Bradley Heppner, who was arrested on charges of securities and wire fraud. Federal agents subsequently seized just over 30 documents from his electronic devices that were generated by the AI tool Claude. The documents were generated using Claude after Heppner received a grand jury subpoena and had hired legal counsel. The documents related to strategies and arguments that Heppner could use in his defense, which were all shared with Heppner’s legal counsel.

Heppner’s attorney argued that the AI generated documents were privileged communications between a client and counsel, as the documents conveyed facts to legal counsel for the purpose of obtaining legal advice. Perhaps key in determining the Court’s ruling, though, was the fact that Heppner generated the documents using AI on his own volition – his counsel never asked him to use AI for any purpose. The government argued that the documents were not privileged because they were not generated for the purpose of obtaining legal advice, and non-privileged documents could not be suddenly privileged simply by sending them to hired counsel.

AI Privilege Ruling

Attorney‑client privilege protects communications (1) between client and attorney, (2) intended to be and kept confidential, (3) for obtaining or providing legal advice. The Court ruled that Heppner’s AI-generated documents failed all three prongs of the privilege test.

Information shared by the defendant (client) to a third party (here, Claude’s AI platform) meant that there was no communication with an attorney. Second, the Court found that Claude’s policies clearly state that any information that users input into the platform may be used for AI-training, and can be disclosed in litigation. As such, Heppner had no reasonable expectation of privacy when he entered information into Claude, nor did the information that he obtained suddenly transform to privileged information simply from sharing the AI-generated information with legal counsel. Finally, since counsel never instructed or suggested that Heppner use an AI tool to obtain information, the use of the Claude platform was not for the purpose of obtaining legal advice.

Key Takeaways

1. Best practice calls for an assumption that utilizing AI platforms does not represent confidential communications, even if the information is later shared with counsel;
2. Information that is privileged can become non-privileged if entered into AI platforms;
3. Carefully consider AI platform policies on information sharing. Even if prompted properly with caveats that the interactions were at the direction of counsel, if the AI platform states that information may be disclosed to other sources, this could destroy any privilege that may exist; and
4. Companies should re-evaluate AI use policies for employees with an eye towards ensuring that employees do not generate or disclose information that could be harmful in litigation.