Nary a day goes by when a client does not contact a lawyer as to a potential or existing legal matter, and tells them that Claude, Grok, or Co-Pilot “told” them the answer to a certain aspect of a legal case, or question whether a certain defense, claim or strategy will solve a legal problem (probably thinking that checking with an AI application will save time and money).
Artificial intelligence platforms such as those above, and others, such as OpenAI’s ChatGPT are now a part of daily life, for distributors and others, and can create a nightmare for clients and their lawyers. It may be the fastest way to discover what a human being facing legal issues – including one acting for his or her employer – truly thinks about the prospects of losing, winning, or the risks in a legal matter.
For opposing counsel to have this information could spell the death knell for a client facing legal challenges, and it may not be protected by the attorney-client privilege or the attorney work-product doctrine – especially if undertaken before an attorney-client relationship is even cemented.
While people in all ways – whether as a business owner, senior executive, employee or potential litigant – may be tempted to turn to AI to “do their own research,” explore potential legal arguments, summarize or draft a document in different ways (e.g., draft a new contract from an existing one that is most favorable to company A and least favorable to company B), or otherwise obtain answers to legal questions, it is critical to understand that the apparent human-like interaction with AI does not confer on AI the legal status of an attorney, where communications and work are confidential, privileged and not subject to disclosure.
United States v. Heppner
A recent first-of-its-kind federal court decision drives home these points. In United States v. Heppner, District Judge Jed S. Rakoff of the U.S. District Court for the Southern District of New York penned a decision that he said “appears to answer a question of first impression nationwide: whether, when a user communicates with a publicly available AI platform in connection with a legal matter” – there a pending criminal investigation – “are the AI user’s communications protected by attorney-client privilege or the attorney work product doctrine?”
Judge Rakoff’s answer was a broadly applicable “no.”
Briefly, the underlying case involved securities fraud, wire fraud, and related charges brought by the U.S. Department of Justice against the defendant, Heppner, where 31 documents seized by federal agents involved Heppner’s communications with Claude.
The communications were particularly significant because they occurred after Heppner had received a grand jury subpoena and learned that he was the subject of a criminal investigation. Without being instructed by counsel to do so, Heppner had printed reports generated by Claude addressing potential defense strategies and arguments concerning the facts and law that the government might raise.
Heppner subsequently claimed those documents were protected by both the attorney-client privilege and the work product doctrine, to later provide to his counsel to obtain legal advice, but the Court rejected those arguments.
An AI Platform Is Not an Attorney, and Its Communications Are Not Confidential
The attorney-client privilege protects certain confidential communications between an attorney and client made for the purpose of obtaining or providing legal advice. It is a narrowly construed protection rather than a blanket shield covering anything related to a legal matter.
Judge Rakoff concluded that Heppner’s communications with Claude failed to satisfy at least two, and potentially all three, of the requirements necessary for attorney-client privilege:
- Claude is not an attorney. The communications at issue were between Heppner and an AI platform, not between Heppner and his lawyer. As the court recognized, Heppner could not claim that Claude was an attorney, so there was no attorney-client relationship between the defendant and the AI platform.
- The communications lacked the necessary confidentiality. Anthropic’s privacy policies informed users that personal information could be disclosed to third parties in connection with claims, disputes, or litigation. As such, the court determined that Heppner could not establish a reasonable expectation that his communications with Claude would remain confidential.
- The communications were not made for the purpose of obtaining legal advice from an attorney. Heppner did not use Claude at the direction of counsel; and Claude itself disclaims providing legal advice. The fact that Heppner later gave the AI-generated documents to his attorney did not retroactively transform those communications into privileged communications.
That last point is particularly important. Simply providing an AI-generated document to a lawyer (or even having it in your possession if later sought in discovery) does not necessarily make the underlying interaction with the AI platform privileged. That issue is a function of many facts and circumstances. The Heppner decision serves as an important warning for business owners, executives, and employees who may be tempted to use AI as a substitute for consulting legal counsel, or even to “prime the pump” on resolving a legal issue; worse, should AI generated material relating to a legal matter be subject to discovery, the damage could be irreparable; essentially case ending material.
AI-Generated Materials Also Did Not Qualify as Attorney Work Product
Heppner also attempted to invoke the work product doctrine, which at its core is to protect the mental impressions and/or opinions of counsel, and sometimes can protect certain material prepared by counsel or at his/her direction – such as opinions of experts and/or investigative material – but only in anticipation of litigation or for use at trial. Depending on the jurisdiction, the attorney work product doctrine is typically a narrow privilege.
Judge Rakoff also found that protection did not apply to Heppner’s communications with Claude. He determined that the materials were not prepared by counsel or at his or her direction. Nor did the communications reveal or embody the defense strategy developed by Heppner’s attorneys. Instead, Heppner independently sought information from the AI platform without being instructed to do so by counsel.
The fact that a user intends to discuss the AI-generated material with an attorney later in the lineage of a legal matter or claims to intend to use it in anticipation of a litigation dispute does not necessarily change its unprotected character and/or that it cannot be had by your opponent in a legal dispute clearly destined for potential or in actual litigation.
AI’s Novelty Does Not Change Established Legal Principles
While the Heppner case arose in the context of a criminal investigation, its reasoning has potentially significant implications in civil litigation. Caution is required, and even senior executives can be “tripped up” sharing their truth of a legal dispute with an AI platform. Providing sensitive information, legal or litigation related facts, potential legal strategies or simple inquiries into a publicly available AI platform is a potential recipe for disaster. For distributors involved in a legal matter that could lead to and certainly matter in actual or potential litigation, err on the side of caution and work around creating non privileged material that could be damaging to the outcome of a legal matter.
The law is always playing catch-up with new technological advancements, and AI is no different. This case will undoubtedly not be the last to apply existing legal frameworks to artificial intelligence. But as Judge Rakoff concluded, “AI’s novelty does not mean that its use is not subject to longstanding legal principles, such as those governing the attorney-client privilege and the work product doctrine.”
If you have any questions or concerns related to the legal implications of your company’s use of generative AI technology, please contact me at 312-840-7004 or fmendelsohn@burkelaw.com.
