“What antitrust concerns could our regulators bring up if tomorrow’s merger moves forward?” This is the question a CFO asks AI right before a major board meeting. Little does this well-meaning executive realize the legal exposure her query may have just created for her and her company.
To appreciate how and why, let’s back up to discuss a recent ruling sending shockwaves through business and legal circles. It pertains to a case involving Bradley Heppner. In October of last year, a federal grand jury indicted the Texas financial executive who once ran GWG Holdings, a publicly traded company. Heppner was accused of defrauding investors to the tune of $150 million through self-dealing transactions across two companies clandestinely under his control.
Here’s where AI comes in. As the law firm Venable explains, “After receiving a grand jury subpoena and retaining counsel, Heppner used Claude to prepare reports outlining potential defense strategies and legal arguments. He later shared those reports with his attorneys and asserted privilege over them.”
Upon searching his home, the FBI seized Heppner’s AI-generated files along with his devices. When his counsel tried to assert attorney-client privilege and work product to keep the documents from the prosecutor, Judge Jed S. Rakoff struck down their assertion in United States v. Heppner, ruling from the bench for the government. Judge Rakoff’s rationale was that the AI documents lacked privilege for at least “two if not all three elements of the attorney-client privilege,” according to the Feb. 17, 2026, memorandum opinion.
These are as follows:
- It found that there was no attorney-client relationship with Claude.
- There was no reasonable expectation of confidentiality in using Claude.
- Heppner wasn’t communicating with Claude for the purpose of obtaining legal advice, as Claude itself disclaims any ability to provide it.
Returning to our CFO in the above story, a related problem to this case could arise for her and her company, one that every business leader should be aware of. Anyone who consults AI is creating a discoverable document that could be used against them in court. In the above instance, should an antitrust case ensue from the merger, opposing counsel could subpoena OpenAI to obtain that information in discovery and leverage it against her.
To understand what’s involved in United States v. Heppner and its implications for the AI Age, I spoke to Simon Khinda, a partner at Bridge Law LLP who heads up the firm’s transactional and international business practice. He told me something surprising. “Even though this case had at its foundation a discussion on the use of AI, ultimately the law that was applied was not unique because it was AI. It went to the core of attorney-client privilege and confidentiality.”
Per Khinda, Heppner had more to do with well-established principles of law than AI per se. It concerned three longstanding principles that must be considered to enjoy the legal protection the defendant’s counsel sought.
- Attorney-client privilege only exists if there is an actual attorney-client relationship.
- Communications must be made at the direction of counsel.
- Confidentiality must be preserved.
By this logic, should anyone break any one of these elements, assumed privilege goes away. Per Khinda’s explanation, these perennial considerations were what Judge Rakoff was responding to. Though the AI angle has received much attention, Heppner was an open-and-shut application of pure legal precedent.
Interesting as this legal development is, there is another dimension to the evolving AI story. The same day Judge Rakoff ruled, Magistrate Judge Anthony P. Patti of the Eastern District of Michigan seemingly came to the opposite conclusion in Warner v. Gilbarco, Inc. In that case, a pro se plaintiff used ChatGPT while bringing employment discrimination claims against her former employer. Judge Patti denied the defendants’ motion to compel discovery of all materials pertaining to such AI assistance.
As Perkins Coie explains, “The decision illustrates the other side of the coin: Where the facts support the elements of the work product doctrine under well-settled law, generative AI tools do not disturb those protections.”
At first glance, Warner appears to contradict Heppner. Instead, Khinda frames it as a mirror image. Heppner failed because the elements of attorney-client privilege were never satisfied: there was no lawyer was in the loop, no confidentiality was preserved, and no direction came from counsel.
Warner, on the other hand, succeeded because a different doctrine applied: work product protection, which shields materials prepared in anticipation of litigation from being turned over to the opposing side during discovery. The doctrine differed in Warner, and so did the waiver standard. As we saw with Heppner, attorney-client privilege evaporates the moment a third party is looped in, while work product is only waived by disclosure to an adversary, and an AI tool isn’t an adversary.
According to Khinda, the two rulings come down to age-old principles. What’s novel is that both rulings occurred on the same day, and both involved the hot-button topic of AI.
For now, Heppner and Warner signal how future courts may handle AI in litigation. The good news is that our legal system is not building new AI doctrines even as AI roils markets and the world. “As both Heppner and Warner reveal, levelheaded judges are applying the traditional reasoning they have in the past,” says Khinda. “Courts are asking the reliable questions they’ve always asked: Was a lawyer involved? Was confidentiality preserved? And who had access to the information, and under what terms?”
Such consideration should reassure a public anxious about the AI juggernaut. Then again, it should also unsettle anyone, executive or not, who thinks AI’s novelty will somehow offer them protection in litigation. Building on these cases as recent precedent, we can expect future judges to rely on the same case law adept lawyers have come to expect, no matter how advanced AI grows in the coming years.
Going forward, behavior concerning AI use, not the tool itself, will determine whether sensitive material stays protected or ends up in discovery. Ultimately, the lesson to business leaders everywhere is a human one. Be careful not so much about the AI you use, but rather about how you use it.
