Imagine, for a moment, that you live in a community where a wise old sage offers free counsel and advice regarding any issue and will help solve any problem. For many people, that is exactly what generative AI platforms, like ChatGPT, Gemini, and Claude, have become. These so-called chatbots are engaging, seemingly all-knowing, and incredibly fast in recommending solutions to your every need. Their breadth of access to information and ability to solve complex problems can make them an alluring resource for addressing your most important concerns.
Now, returning to your imaginary community, imagine that, after one particular meeting with the wise old sage, in which you shared and discussed delicate personal information, you later learn that this sage has been chatting it up with other community members, revealing all your most personal, private details. According to one federal judge, that is essentially what happens when you input your otherwise private information into a frontier generative AI model.
United States v. Heppner
Attorney-Client Privilege
Most people are familiar with the concept of attorney-client privilege—the almost sacrosanct protection afforded to communications exchanged between a client and his or her attorney. But, as the Heppner court explains, this privilege is treated very carefully because it can have the effect of excluding otherwise important evidence from trial. This means that only those communications that are intended to be confidential and are in fact kept confidential will be protected. The Honorable Judge Jed Rakoff concluded that Heppner "had no reasonable expectation of confidentiality in his communications with Claude" because Anthropic’s privacy policy openly announces that it uses the input and output data to train itself and that Anthropic reserves the right to disclose such data to third parties. No privacy means no privilege. Even if the information Heppner input into Claude was initially confidential information exchanged between his attorney and him, Heppner compromised that confidentiality and thus waived the privilege the moment he entered it into Claude.
Attorney Work Product
Perhaps less familiar to the general public is the idea of attorney work product. The idea here is that an attorney should be allowed to analyze and prepare his or her client’s case without the threat of his notes and other work product being later used against his client. Again, this protection is narrowly applied by courts and only protects materials created by, or at the direction of, counsel in anticipation of litigation or trial. Heppner fell short of this protection as well. As the court explained, because Heppner’s interactions with Claude were of his own volition, on his own time, and not at the direction of, or in concert with, his attorney, the materials are not protected under the work product doctrine.
Bottom Line
AI platforms like ChatGPT, Gemini, and Claude, are powerful tools and can be incredibly helpful. Indeed, more and more law firms now implement AI solutions in their practice. However, as with most powerful tools, the potential risks involved necessitate caution and oversight. While it is probably tempting to ask the wise old sage for legal advice about your case, save those questions and conversations about your case for your attorney.
Jason Grutter
Partner
Jason Grutter is a distinguished trial attorney who passionately advocates for the rights of beneficiaries and heirs in complex trust and estate disputes.
Author's Bio