If you are working at a law firm implementing AI protocols, you have likely been warned not to share sensitive client/case details on public AI platforms like ChatGPT or Claude because of the potential of waiving attorney-client privilege. While attorney-client privilege is intended to protect all communications between attorney and client involving legal advice or guidance, sharing information on a third-party public platform becomes a risky proposition. Two federal cases recently dealt with this issue and were split in their respective outcomes.

The first case, U.S. v. Heppner, close to home in the Southern District of New York, was a criminal case in which the client, a CEO accused of securities fraud, used AI to find out details about the government’s investigation and then shared the prompts he used and resulting outputs with his attorneys. After the FBI seized those documents, the defense sought to suppress the evidence as privileged. The Court denied the motion, finding that the privilege was waived through third-party disclosure – the third party being the AI software. The Court specifically found that the AI software was an independent entity to which this CEO disclosed sensitive information.

The second case arrived at a different result. In Warner v. Gilbarco out of the Eastern District of Michigan, the Court again grappled with what to do with AI prompts and whether that information shared with a public AI platform, ChatGPT in this instance, waives work-product protection. The Judge here held that sharing prompts with ChatGPT does not waive work-product protection of the attorney and is not discoverable. Specifically, unlike in Heppner, the Judge here found that ChatGPT and the like are tools, not persons, and disclosure to a software tool, as opposed to an adversary or other third party does not waive protection. The Court drew an interesting comparison of AI software to a modern day word processor or legal research database, not an actual third-party recipient. It’s worth noting here that no confidential case information was shared with ChatGPT, although based on the Judge’s decision, it’s unlikely the result would have been different.

Given the differing views of how these courts view and interpret what AI platforms are – a tool vs. independent third-party entity – this seems like it will be a continually evolving situation.

The best guidance at this point is to avoid the risk and not share sensitive case information with any public AI platform. Contact Jeff Basso at jbasso@cmmllp.com for guidance.