AI is Not Your Lawyer: What United States v. Heppner Means for Your AI Legal Searches

For many founders, executives, and creative entrepreneurs, generative artificial intelligence has become an indispensable productivity tool. We use it to draft marketing copy, summarize dense reports, and outline complex strategies. But a recent, first-of-its-kind ruling from the Southern District of New York should give every business owner serious pause before turning to a consumer-grade AI chatbot for anything resembling legal analysis.

In United States v. Heppner, No. 25 CR. 503 (JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026) , Judge Jed S. Rakoff held that thirty-one documents a criminal defendant generated through conversations with a consumer AI platform were protected by neither the attorney-client privilege nor the work-product doctrine. The decision, which Judge Rakoff himself characterized as presenting “a question of first impression nationwide,” has significant implications for how businesses and their counsel must approach generative AI in any context where confidentiality matters.

This post examines the Heppner ruling, contrasts it with a companion decision issued the same day in the Eastern District of Michigan, and outlines what business owners should take away from this emerging body of law at the intersection of AI and emerging technology and traditional civil procedure.

---

Background: The Facts of United States v. Heppner

Bradley Heppner is a Dallas-based financial services executive who was indicted on October 28, 2025 on charges related to securities fraud, wire fraud, and falsifying corporate records arising from his alleged conduct as an executive of GWG Holdings, Inc. and its affiliate, Beneficient. After receiving a grand jury subpoena and retaining defense counsel at Quinn Emanuel, but before his arrest, Heppner began using the consumer version of a publicly available generative AI platform to research legal questions related to the government’s investigation.

Critically, Heppner did this on his own initiative. His attorneys did not instruct him to use the tool, did not supervise his interactions with it, and did not engage the platform as an agent of the legal team. Over the course of these sessions, Heppner inputted information he had learned from his lawyers, prompted the AI to develop defense narratives and analyses, and generated thirty-one documents reflecting that work. He later shared the materials with his defense team.

When the FBI executed a search warrant at Heppner’s residence, agents seized electronic devices containing those documents. The government moved for a ruling that the materials were not privileged. On February 10, 2026, Judge Rakoff granted the motion from the bench. On February 17, 2026, he issued a written opinion explaining the decision in full.

The Court’s Analysis: Attorney-Client Privilege

The attorney-client privilege protects confidential communications between a client and an attorney made for the purpose of obtaining legal advice. Judge Rakoff held that Heppner’s AI-generated documents failed at least two, and likely all three, elements of the privilege.

1. The Communications Were Not Made With an Attorney

The most fundamental flaw was the absence of a lawyer. A generative AI platform is not an attorney; it holds no license, owes no duty of loyalty, and is not bound by the rules of professional responsibility. The court rejected the argument that the privilege could attach simply because Heppner intended to discuss the AI output with his lawyers afterward, observing that “non-privileged communications are not somehow alchemically changed into privileged ones upon being shared with counsel.”

2. The Communications Were Not Made for the Purpose of Obtaining Legal Advice From Counsel

Judge Rakoff also emphasized that the AI tool itself expressly disclaims providing legal advice. A user querying such a system is not seeking counsel’s legal judgment; the user is consulting an information source, much like consulting a search engine or a library treatise. Notes from those independent consultations are not privileged merely because they are later handed to a lawyer.

3. There Was No Reasonable Expectation of Confidentiality

Perhaps most significant for businesses, the court grounded its confidentiality analysis in the platform’s governing terms of service and privacy policy. Those terms disclosed that the operator collects user inputs and outputs, may use that data to train its models, and reserves the right to share data with categories of third parties including regulators. Under those conditions, the court held that Heppner could not have harbored a reasonable expectation that his communications were confidential. This is a meaningful point for any business that handles sensitive client or employee information, and one that intersects directly with broader cybersecurity and data protection obligations.

The Court’s Analysis: Work-Product Doctrine

The work-product doctrine, codified in Federal Rule of Civil Procedure 26(b)(3) and recognized at common law in criminal matters, protects materials prepared in anticipation of litigation by a party, an attorney, or an agent of either acting at counsel’s direction.

Heppner’s documents failed this test as well. Defense counsel conceded that Heppner had created the materials “of his own volition” and that the legal team did not direct him to run the AI queries. Without attorney direction, the AI platform could not be characterized as an agent of counsel, and the resulting documents did not reflect defense counsel’s litigation strategy. Judge Rakoff suggested the analysis might have come out differently had counsel directed the use of the tool, though commentators have noted that this caveat raises its own practical and doctrinal complications.

The Companion Case: Warner v. Gilbarco

On the same day Judge Rakoff issued his bench ruling, Magistrate Judge Anthony P. Patti of the Eastern District of Michigan reached the opposite conclusion in Warner v. Gilbarco, Inc., No. 2:24-cv-12333. There, a pro se employment-discrimination plaintiff had used a consumer AI platform to assist in preparing litigation materials. The defendants moved to compel production of those materials, arguing that uploading information to the AI waived any protection.

Judge Patti denied the motion. He reasoned that generative AI platforms are “tools, not persons,” and that a holding to the contrary “would nullify work-product protection in nearly every modern drafting environment.” Because the pro se plaintiff was effectively acting as her own counsel and prepared the materials in anticipation of litigation, the documents qualified as work product. And because Sixth Circuit law requires disclosure “to an adversary or in a manner likely to reach an adversary’s hands” to constitute waiver, simply inputting information into the AI did not waive the protection.

Read together, Heppner and Warner do not represent a doctrinal split so much as a fact-driven application of well-settled principles to a new technology. The lesson is not that AI categorically destroys privilege or work product; it is that long-standing rules about confidentiality, third-party disclosure, and attorney direction apply with full force when the third party is a chatbot.

Why Heppner Matters for Business Owners

Most business owners will never face a federal indictment. But the principles articulated in Heppner apply with equal force to civil litigation, regulatory inquiries, internal investigations, employment disputes, and intellectual property matters. Any context in which a business communicates sensitive information that might later become the subject of discovery is a context in which Heppner is relevant.

Consider the founder who, facing a business litigation matter, uses a consumer AI tool to draft a narrative of “what really happened.” Or the executive engaged in a commercial litigation dispute who pastes internal correspondence into a chatbot and asks it to identify weaknesses in the company’s position. Or the partner in a closely held LLC who uses an AI assistant to model settlement scenarios. In each case, the resulting documents may well be discoverable. Worse, depending on the information shared with the tool, the underlying privileged communications between client and counsel may themselves be at risk of waiver. As the government argued and Judge Rakoff agreed in dicta, sharing privileged information with a third-party AI platform may constitute a waiver of the privilege over the original attorney-client communications.

The exposure is particularly acute for matters involving information that the business has independently committed to protect, such as data subject to NDA and confidentiality agreements. Inputting that information into a consumer AI tool may create a separate breach of contract problem in addition to any privilege concerns.

Practical Guidance: Using Generative AI Without Forfeiting Protection

The takeaway from Heppner is not that businesses should avoid generative AI. The technology is too useful, and the productivity gains too real, for blanket avoidance to be a serious strategy. The takeaway is that businesses must use generative AI thoughtfully, particularly where confidentiality, privilege, or work product are at stake.

Recommended Practices

• Use generative AI freely for non-sensitive tasks. Drafting marketing copy, summarizing public articles, brainstorming names, generating outlines, and similar tasks pose no privilege concern.
• Review the terms of service of any AI tool before inputting sensitive information. If the tool reserves the right to train on user data or share it with third parties, treat the platform as public.
• Adopt written internal policies and procedures governing AI use. Clear written policies establish expectations for employees, document the company’s diligence, and create a paper trail that supports work-product and confidentiality claims if a dispute arises.
• Consult counsel before using AI in any matter involving actual or anticipated litigation, regulatory exposure, or sensitive negotiation. Use of an AI tool in those contexts should be directed and supervised by counsel, not undertaken on the client’s initiative.
• Consider enterprise-grade tools with appropriate confidentiality protections. Many AI platforms now offer enterprise tiers with contractual commitments not to train on user inputs, and those terms materially change the confidentiality analysis.
• Document the engagement. Where counsel directs the use of an AI tool, that direction should be memorialized in writing.

Practices to Avoid

• Do not input trade secrets, privileged communications, or sensitive litigation strategy into a consumer AI platform. Once inputted, the information is, at best, outside your control.
• Do not assume that deleting a chat removes the data. Most platforms retain user data on their servers regardless of what is visible to the user.
• Do not generate “defense narratives” or self-assessments of a dispute through a consumer AI tool. That is precisely the conduct the Heppner court refused to protect.
• Do not assume that sharing AI-generated work product with counsel after the fact will retroactively confer privilege. It will not.

How Isaboke Law Firm Approaches This Issue

At Isaboke Law Firm, PLLC, we view generative AI as a tool to be integrated carefully, not avoided reflexively. For our fractional general counsel clients, we help establish AI-use policies that protect privilege while preserving the productivity benefits of the technology. For litigation defense and trademark clients, we directly supervise any AI-assisted work product where confidentiality matters, ensuring that the work is performed under conditions that the Heppner court would recognize as protected.

If your business is currently using generative AI in connection with a contract dispute, an employment matter, an IP issue, or any other sensitive context, and you are not certain whether your communications and work product are protected, that uncertainty itself is a problem worth addressing. We are happy to talk through your current practices and identify any exposure.

Conclusion

United States v. Heppner does not announce a new rule unique to artificial intelligence. It applies the traditional elements of attorney-client privilege and work-product doctrine to a defendant who treated a consumer chatbot as a stand-in for legal counsel. The result was predictable in retrospect: a tool that is not a lawyer, that is governed by a privacy policy authorizing third-party data sharing, and that is consulted without attorney direction does not produce privileged or protected materials.

For business owners, the practical message is straightforward. Generative AI is a powerful productivity tool. It is not a substitute for counsel, and treating it as one can have serious consequences in litigation. The careful integration of AI into a legal workflow, directed and supervised by counsel and supported by appropriate enterprise tooling, remains entirely viable. The unsupervised consumer use of AI for matters that may someday end up in discovery does not.