Spotlight: The AI Privilege Split—TWO FEDERAL COURTS, TWO OPPOSITE ANSWERS, AND WHAT EXIT-READY FOUNDERS NEED TO KNOW
- Clay Chamberlain
- Feb 27
- 6 min read
Read to discover how to:
|
Part 1: The $300 Million AI Trap Part 2: The Ruling Is In — Heppner Update
TWO COURTS. TWO ANSWERS. ONE QUESTION THAT AFFECTS EVERY FOUNDER PREPARING FOR AN EXIT.
If you've been following this series, you know the story so far. In Part 1, we broke down the Heppner case: a founder who used AI to research his own legal situation and watched prosecutors demand every prompt he ever typed. In Part 2, Judge Rakoff in the Southern District of New York ruled those AI-generated documents were protected by neither attorney-client privilege nor the work product doctrine. The message was stark: your AI research can be used against you.
That was two weeks ago.
Now a federal court in Michigan has reached the opposite conclusion.
And the implications for exit-ready founders just got significantly more complicated.
WHAT JUST HAPPENED IN WARNER V. GILBARCO
On February 10, 2026, Magistrate Judge Anthony P. Patti of the Eastern District of Michigan issued a ruling in Warner v. Gilbarco, Inc., Case No. 2:24-cv-12333, that directly collides with the Heppner decision.
The facts are different, this is an employment dispute, not a securities fraud prosecution, but the legal question is identical: can the opposing party force you to hand over your AI-generated research?
Sohyon Warner, a pro se plaintiff suing her former employer for employment discrimination, used ChatGPT to assist with her litigation. The defendants, Gilbarco and its parent company Vontier, filed a motion to compel production of all documents related to her use of AI tools. They wanted every prompt, every output, every conversation she had with the platform.
The court said no. Emphatically.
THE FIVE PILLARS THAT HELD
Judge Patti's ruling rested on reasoning that reads like a point-by-point rebuttal of the Heppner decision even though the court never cited it directly.
AI research prepared for litigation is work product. The court applied Federal Rule of Civil Procedure 26(b)(3)(A) without hesitation. Materials prepared in anticipation of litigation, even materials created by a party acting without an attorney, are protected. Warner was preparing for her lawsuit. Her AI research was part of that preparation. End of analysis.
Pro se litigants have the same protection. The court cited its own prior precedent to confirm that a self-represented party has every right to assert work product protection over litigation preparation materials. You don't need a law license for the doctrine to apply.
AI tools are tools, not adversaries. This is the line that matters most. The court held that using ChatGPT does not waive work product protection because, under Sixth Circuit precedent, waiver requires disclosure to an adversary or in a way likely to reach an adversary's hands. And as the court put it directly: AI programs are "tools, not persons, even if they may have administrators somewhere in the background."
Compelling AI research invades mental processes. The court agreed with Warner's argument that the defendants' motion was really asking for her internal analysis and thought process "rather than any existing document or evidence, which is not discoverable as a matter of law." Forcing disclosure of how someone uses AI to prepare their case is functionally the same as forcing disclosure of their litigation strategy.
The slippery slope argument won. Judge Patti adopted Warner's position that the defendants' theory "would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed." If using AI to research and draft waives protection, then using spell-check, legal research databases, or document drafting software would too. The court refused to open that door.
The ruling ended with a line that could have been written for this series: the court called the defendants' pursuit of AI disclosure "a distraction from the merits of this case" and ordered that their "preoccupation with Plaintiff's use of AI needs to abate."
Want the implementation tool?
This post is based on the NDA Control Toolkit from the Locking in Leverage Masterclass. Subscribers to Big Exit Insiders get it free.
THE COLLISION COURSE — HEPPNER VS. WARNER
So here we are. Two federal courts. Same question. Opposite answers.
In the Southern District of New York, Judge Rakoff ruled that Bradley Heppner's 31 AI-generated documents were fair game for prosecutors. No privilege. No work product protection. The reasoning: AI is not an attorney, confidentiality was destroyed when Heppner typed his first prompt into a commercial platform, and his lawyers didn't direct the research.
In the Eastern District of Michigan, Judge Patti ruled that Sohyon Warner's AI research is fully protected work product. The reasoning: AI is a tool, not a person. Using it doesn't waive anything. And compelling its disclosure would invade the mental processes that the work product doctrine was designed to protect.
These courts aren't just disagreeing on the margins. They are operating from fundamentally different premises about what AI is and what using it means.
Heppner treats AI as a third party: a disclosure recipient that destroys confidentiality. Warner treats AI as a tool: no different from a calculator, a legal research database, or a word processor.
Both courts are applying established doctrine. Both cite respected precedent. And until an appellate court weighs in, both rulings stand in their respective jurisdictions.
This is what lawyers call a split. And when there's a split, the only guaranteed outcome is uncertainty.
WHY THIS MATTERS MORE THAN YOU THINK
If you're an exit-ready founder preparing for an 8-figure transaction, you might be tempted to read this as a legal technicality that doesn't affect you. You're not in litigation. You're not facing a federal indictment. You're just researching deal terms and running financial scenarios.
But here's the thing about uncertainty in the law: it doesn't protect anyone. It exposes everyone.
Consider the scenario that plays out after almost every major transaction: a post-closing indemnification dispute. The buyer claims you breached a representation. You claim you didn't. Discovery begins. And somewhere in that discovery process, someone asks for your pre-transaction research: the AI conversations where you analyzed your own financials, explored your own exposure, tested your own negotiating positions.
Under Heppner, that research is discoverable. Under Warner, it's protected. And you won't know which rule applies until a judge in your jurisdiction decides possibly years after you typed the prompt.
That is not a risk any sophisticated founder should accept. Not when the stakes are this high. Not when the fix is this simple.
WHAT THE SPLIT ACTUALLY TEACHES FOUNDERS
Strip away the legal complexity and both rulings point to the same practical truth. The one we've been saying since Part 1 of this series.
The protection depends entirely on the professional framework surrounding the AI use.
In Heppner, the founder acted alone. No attorney direction. No professional structure. No protective architecture. The AI research floated free, untethered to any legal relationship and the court treated it accordingly.
In Warner, the litigant used AI as part of her litigation preparation, a recognized legal activity with established protections. The court saw the AI as an extension of her preparation process, not a separate disclosure event.
The difference isn't the AI platform. It's not the type of case. It's not the jurisdiction, though that matters too. The difference is whether the AI use sits inside or outside a recognized professional framework.
For exit-ready founders, that framework is the attorney-client relationship.
THREE THINGS THAT HAVEN'T CHANGED
The legal landscape is shifting fast, but the core guidance from this series remains rock-solid and Warner actually reinforces it.
First, engage your M&A counsel before you start preparing. This was true after Heppner. It's doubly true now. When your attorney directs your AI research—when the prompts, the analysis, the strategic thinking all happen within the scope of legal representation—you're protected under both the Heppner framework and the Warner framework. Attorney direction is the common denominator that wins in every courtroom.
Second, assume the worst-case standard applies until the law settles. Right now, the Heppner standard is the more restrictive one. If you treat every unsupervised AI prompt as potentially discoverable, you will never be caught off guard regardless of which jurisdiction your dispute lands in.
Third, the sacred flaw still applies. The instinct to research independently and to figure it out yourself before calling the lawyer. This is the same instinct that built your company. It's also the instinct that creates unprotected material during the most sensitive period of your business life. Warner's AI research was protected because it was tied to a formal legal proceeding. Your informal, pre-engagement AI exploration of deal terms? That's Heppner territory until your attorney is involved.
MAIN TAKEAWAY
Two federal courts have now drawn opposite lines on whether AI research is protected and until an appellate court resolves the split, the only way to guarantee your pre-transaction preparation stays confidential is to make sure your M&A attorney is directing the work from the very first prompt.
Become a Big Exit Insider
Get the NDA Control Toolkit when you subscribe. Plus: biweekly deal intelligence, early framework releases, and priority access to the M&A Legal Masterclass.
Comments