Litigants trying to understand their legal situation with the help of AI are either totally fine or totally screwed. Welcome to the modern practice of law!
Earlier this month, Judge Jed Rakoff of the Southern District of New York ruled in United States v. Heppner that 31 documents that a criminal defendant generated using the consumer version of Anthropic’s Claude were not protected by attorney-client privilege or the work product doctrine. Meanwhile, Magistrate Judge Anthony P. Patti of the Eastern District of Michigan heard a substantially similar discovery dispute and concluded in Warner v. Gilbarco, Inc. that of course the other side can’t seize the litigant’s legal work just because it went through a large language model.
In Heppner, the defendant had already engaged counsel and queried the AI on his own to prepare materials for a meeting with his lawyers. By contrast, the party in Warner represented herself and used AI to prepare her own case. The fact that Warner acted as her own counsel and the searches directly reflect her legal strategy goes a ways toward explaining the distinction, but it doesn’t go quite far enough.
The Heppner decision talked about AI as a non-lawyer third-party whose terms of service acknowledge that inputs may not remain confidential. Those issues don’t change just because the party is acting as their own counsel.
Judge Rakoff identified a Claude ping as a third-party disclosure. Judge Patti drew a distinction, based on the D.C. Circuit in United States v. American Telephone & Telegraph Co., that voluntary disclosure to a third party does not, by itself, waive work product protection. To defeat the work product doctrine, Judge Patti ruled, the party has to disclose the material directly to an adversary or in some way likely to reach the adversary’s hands. So unless you’re litigating against Anthropic, you would be fine.
That’s where Judge Rakoff’s opinion holds to the letter of the law in a way that undermines the spirit in a world of AI tools. The Heppner confidentiality analysis pointed to Anthropic’s privacy policy and found no reasonable expectation of confidentiality, because the company asserts that it can collect user data, train models on it, and disclose information to government authorities and third parties. Therefore, Rakoff reasoned, sharing information with Claude is like discussing your legal strategy in a crowded room.
Except every major cloud service has substantially identical terms. If the client saves emails and documents on Microsoft OneDrive or something, have they waived all protections? If the client uses Gmail, they arguably waive privilege under this reasoning. The Heppner analysis makes sense in the abstract, but practically we can’t allow our new cloud-based reality to obviate traditional protections.
And that’s if you think an AI product is a third-person at all, a concept that Judge Patti wasn’t sold on:
ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.
Had Heppner taken information that he received from his attorneys and gone to the local law library or even run standard Google searches, we wouldn’t be having this discussion. But these days, Google pumps your searches into its AI anyway… does that make a client’s internet search to figure out that legalese the lawyer just said on the call presumptively discoverable? That can’t be right. It gets even worse when you realize CoPilot is baked into Microsoft Office and Google’s Gemini is embedded in Workspace. The notes a client takes of an attorney meeting are traditionally protected, but if boilerplate terms of service for cloud applications can defeat the expectation of privacy, all bets are off.
These are previously untested applications of rules that were pretty clear before running aground on the jagged rocks of technology. As the Wagner opinion notes:
Additionally, the Court agrees with Plaintiff that the pursuit of this information is “a distraction from the merits of this case[,]” and that Defendants’ theory, which is supported by no case law but only a Law360 article posing rhetorical questions “would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed.”
A Law360 article? If it were an Above the Law article maybe, but come on.
It’s worth noting, as Jennifer Ellis observed, Judge Patti handles discovery disputes every day and has a more intimate experience with the ways technology plays hell with the letter of the law. Judge Rakoff doesn’t spend his whole day on these complications. As these cases proliferate, expect to see a divide between the magistrate judges and the district judges.
But Heppner is, for the time being, the go-to standard of the most important federal court in the country. Every Biglaw firm has already blasted out a client alert ruminating on its implications. Clients interested in AI are advised to “use enterprise tools,” though that’s unlikely to resolve the underlying problem. Unless (or until) the AI bubble bursts so spectacularly that we’re back to our tried and true tools, the question remains whether courts should treat AI chat history as the equivalent of shouting a legal strategy in Times Square.
For now, the proper advice is that clients shouldn’t risk talking about their cases with AI. And maybe save everything locally. And maybe don’t run internet searches with Gemini. You know what? Maybe just don’t use a computer at all.
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news. Joe also serves as a Managing Director at RPN Executive Search.
The post Does Asking ChatGPT A Legal Question Make It Discoverable? It Depends! appeared first on Above the Law.