Written by Charles Dresser

​Generative AI tools are pervasive. Tools specialized for many domains are commonly available and widely used. The emergence of these tools has been so precipitous that the relationship between generative AI products and the law is only beginning to be explored. Business leaders should be generally aware of the AI environment, its ostensible use, its internal use of received data and privacy policy agreements required of end users.

Pulling from my professional experience as a partner at a technology law firm, ​in this article, I will discuss the emerging definitions of the relationships between AI products and the policy of privilege utilized in the discoverability of documents in legal cases.​​

Most business leaders know not to write anything in an email that they would not want to see in a court room. Likewise, AI-based exploration of sensitive ideas​, if relevant to a lawsuit, will likely be disclosed with litigants in discovery. But there are legal protections that could prevent your chat history from becoming “Exhibit A.”​

​Two Sources Of Protection Documents

When confronted with motions to gather documents created by a legal adversary, the court must decide whether sought documents are protected. Two sources of protection of documents are attorney-client privilege and work product protection.​

Attorney-client privilege is easily understood as the protection of communication between a trusted (human) attorney and the person seeking legal advice. The content of that communication is protected and cannot be gathered for consideration in court proceedings. However, when that communication is shared outside the relationship (with a third-party), attorney-client privilege is waived. ​

Work product protection is afforded to certain ideas, notes or documents that an attorney has regarding a current or future litigation. Any information that suggests how the attorney is thinking about the case (i.e., mental impressions) is protected as work product and is not discoverable.​

Examples

Two recent cases give us a glimpse at how these two sources of privilege can be construed when a generative AI tool is inserted into the mix.​

Warner v Gilbarco, Inc. is an employment-discrimination case in which the plaintiff, Sohyon Warner, claimed her former employer, Gilbarco, Inc., had based her termination on race and gender discrimination in contravention of federal and state statutes. As is increasingly common in business, Warner had used AI to help prepare for the suit. And the defendant, Gilbarco, in a pre-trial motion, sought production and presentation of all AI-related materials relevant to the case that had been produced, in anticipation of litigation, by Warner with AI tools arguing such use waived any attorney-client or work product protections.

Judge Drain denied the motion, citing the work product doctrine. Fed. R. Civ. P. 26(b)(3) (“[o]rdinarily, a party may not discover documents and tangible things that are prepared in anticipation of litigation or for trial by or for another party or its representative”). Notably, the plaintiff, Warner, was defending herself pro se, without an attorney’s representation. In denying the motion, Judge Drain noted that generative AI programs are tools, not people, so attorney-client privilege was not necessarily waived by use of generative AI, and such communication should not “in itself” suffice for waiver of the work product privilege.​

In noting that AI programs are tools and not people, Drain implied that privacy rights had not been waived by Warner’s sharing of insights with the AI agent and further stated that the communication had not resulted in waiver of the work product privilege, either.​

However, another court found differently in U.S. v Bradley Heppner when the accused, Heppner, also not an attorney, sought to have his communications with an AI agent protected.​

Heppner’s counsel argued that FBI-seized exchanges Heppner had with generative AI applications should be protected by attorney-client privilege and by work product protection. The court ruled that such exchanges were protected by neither attorney/client privilege nor work product protection. Heppner argued that he had prepared his AI documents in anticipation of discussions with his attorney, had used ideas provided by his attorney and used them in discussions with his attorney. The Government moved for a ruling that Heppner’s documents were protected by neither attorney-client privilege nor the work product doctrine.

The Government’s motion was granted permitting access to Heppner’s incriminating conversations with an AI chatbot. In granting the motion, the court considered that Heppner had not had privileged communication with an attorney; he communicated with software, Claude. Not only were Heppner’s conversations not within a trusting human relationship, they were also not private. By contract, Claude’s owner, Anthropic, uses such communications for training Claude, and retains the right to share input as it sees fit. Hence, Heppner did not have an attorney-client relationship for protection or even a reasonable expectation of privacy.

Additionally, the work product doctrine was not applicable since the AI user, Heppner, was not an attorney. Importantly, Heppner’s attorney did not order or direct Heppner use a generative AI tool in anticipation of litigation.

Had Heppner been advised by his attorney to interact with the AI agent, his interactions with Claude may have received work product protections, but that was not the case. The documents produced by Heppner were deemed not to have warranted protection under the tightly construed work product doctrine because they were not produced by counsel, nor by Heppner acting at the behest of counsel.​

​What Can We Learn From This?

These two cases provide a look at how the courts perceive attorney-client privilege and work product doctrine in cases in which generative AI has been used. Though AI is an increasingly important tool for business leaders, those leaders should be cognizant that the generative AI use might be viewed by the courts, should they face related litigation.​

Business leaders should know that AI tools used according to privacy policies that permit third-party disclosure are likely to waive attorney client privilege. Likewise, communication with an AI tool is unlikely to receive attorney client privilege protections, without a trusted human relationship. However, AI tool interactions either by counsel, or at the direction of counsel, may receive protection.