AI Attorney-Client Privilege: Lessons from Heppner

As generative artificial intelligence becomes more common, many people now use AI platforms to help organize thoughts, research legal issues, draft legal documents, or better understand a business dispute. A recent federal court decision, however, is an important reminder that communications with AI are not the same as communications with an attorney.

In United States v. Heppner, the court addressed what it described as a likely issue of first impression: whether communications between a criminal defendant and a publicly available generative AI platform were protected by attorney-client privilege or the work-product doctrine. The court held that they were not.

The decision is important because it shows how traditional confidentiality rules may apply when a person uses AI to discuss sensitive legal issues. For individuals, business owners, executives, and professionals, the lesson is straightforward: do not assume that information entered into a public AI platform will be treated as private or privileged.

Background of the Heppner Case

Mr. Heppner was indicted in October 2025 on securities fraud, wire fraud, conspiracy, false statements to auditors, and falsification of corporate records arising from his alleged conduct involving GWG Holdings, Inc. and affiliated entities.

After receiving a grand jury subpoena and learning that he was the target of the Government’s investigation, Heppner independently used Claude, a generative AI platform operated by Anthropic. He used the platform to generate written analyses about possible defense strategies, legal theories, anticipated charges, and factual arguments.

Following Heppner’s arrest, the FBI executed a search warrant at his home. During that search, agents seized approximately thirty-one documents reflecting Heppner’s communications with Claude.

Heppner’s attorneys argued that the documents should be protected by attorney-client privilege and the work-product doctrine. They claimed the materials were created in anticipation of commercial litigation, included information Heppner had learned from counsel, were intended to be used in discussions with counsel, and were later shared with counsel. The court rejected those arguments.

The Federal Court’s Main Question

The central issue was whether communications with a public AI platform could be treated like confidential communications with an attorney or attorney-directed litigation materials. The court concluded that they could not.

The court’s reasoning focused on two important legal protections: attorney-client privilege and the work-product doctrine. Although these doctrines are related, they protect different things. Attorney-client privilege generally protects confidential legal communications between a client and lawyer. The work-product doctrine generally protects certain materials prepared by attorneys, or at their direction, in anticipation of litigation. In Heppner’s case, the court found that neither protection applied.

Why Attorney-Client Privilege Did Not Apply

Attorney-client privilege generally protects confidential communications between a client and an attorney when the purpose of the communication is to obtain or provide legal advice. The court found that Heppner’s communications with Claude failed several parts of that test.

First, Claude was not an attorney. The court emphasized that attorney-client privilege depends on a relationship between a client and a licensed professional who owes ethical and legal duties to the client. A generative AI platform cannot form that kind of attorney-client relationship.

A simple example helps explain the point. If a business owner discusses a potential breach of contract lawsuit directly with her lawyer, that conversation may be privileged. But if she enters the same facts into a public AI chatbot and asks it to suggest claims, defenses, or litigation strategy, she is not communicating with a lawyer. She is communicating with a third-party technology platform. That difference mattered to the court.

Communications With AI Were Not Confidential

The court also found that Heppner lacked a reasonable expectation of confidentiality when he communicated with Claude. The court relied heavily on Anthropic’s privacy policy, which informed users that Anthropic could retain user inputs and outputs, use communications to train Claude, and disclose data to third parties, including governmental authorities or in connection with business litigation or legal disputes.

Because Heppner voluntarily disclosed information to a publicly available AI platform that could retain and potentially share those communications, the court held that the necessary confidentiality for attorney-client privilege was missing.

This is a critical takeaway. Privilege usually depends on keeping the communication confidential. Once sensitive information is voluntarily shared with a third party, privilege may be waived or may never attach in the first place.

For example, suppose a company executive writes private notes about a contract dispute with a vendor and shares those notes only with the company’s attorney. Those notes may remain confidential. But if the executive first uploads the same information into a public AI platform and asks for legal strategy, a court may find that confidentiality was already lost before the attorney ever saw the document.

Claude Was Not Providing Legal Advice

The court also rejected the argument that Heppner was using Claude to obtain legal advice. Although Heppner may have intended to later discuss the AI-generated materials with his attorneys, the court explained that the relevant question was whether Heppner was seeking legal advice from Claude at the time he used the platform.

The court found that he was not seeking legal advice from an attorney. He was interacting with an AI platform that expressly disclaimed providing legal advice and advised users to consult qualified attorneys. The court noted that the analysis might have been different if Heppner’s lawyers had specifically directed him to use Claude as part of their representation. In some circumstances, third parties assisting attorneys may fall within privilege protections. But here, defense counsel admitted that they never instructed Heppner to use Claude. He acted entirely on his own.

Later Sharing the AI Materials With Counsel Did Not Make Them Privileged

The court also made another important point: a non-privileged document does not become privileged simply because it is later sent to an attorney. Heppner argued that he intended to use the AI-generated materials in discussions with his legal team and eventually shared them with counsel. The court was not persuaded.

The key issue was whether the documents were privileged when they were created. Because Heppner created them through his own communications with a third-party AI platform, outside the direction of counsel, the later transfer to his attorneys did not transform them into privileged materials.

This is a common misconception. A person cannot generally take an otherwise non-confidential document, forward it to a lawyer, and then claim the document is protected from disclosure simply because a lawyer received it.

Why the Work-Product Doctrine Did Not Apply

The court separately considered whether the AI documents were protected under the work-product doctrine. The work-product doctrine is designed to protect materials prepared by attorneys, or at the direction of attorneys, in anticipation of litigation. It is especially concerned with protecting attorney strategy, legal theories, mental impressions, and litigation planning.

Although Heppner created the documents while anticipating an indictment and litigation, the court found that this was not enough. The documents were not prepared by his attorneys. They were not prepared at his attorneys’ direction. They did not reflect counsel’s mental impressions or litigation strategy at the time they were created. Instead, the documents were created entirely by Heppner through his independent use of Claude.

For example, if an attorney asks a paralegal or consultant to help prepare litigation research for a pending lawsuit, that work may be protected because it was done at the direction of counsel. But if a client independently uses a public AI chatbot to brainstorm defenses before being instructed by counsel, a court may treat those materials very differently.

The Court’s View of AI and Traditional Privilege Rules

The court framed the case as an important early decision involving generative AI and long-standing privilege principles. The key point was that AI’s novelty does not change the basic rules. Attorney-client privilege still requires a confidential communication with an attorney for the purpose of legal advice. The work-product doctrine still focuses heavily on attorney-directed litigation preparation and attorney strategy.

Because Heppner voluntarily communicated with a third-party AI platform, without attorney direction and without a reasonable expectation of confidentiality, the court held that the documents were not protected.

Practical Takeaways From the Heppner Decision

The most important takeaway from United States v. Heppner is that communications with public AI platforms should not be treated as private legal communications. Individuals and businesses should be careful before entering sensitive facts, litigation strategy, admissions, timelines, internal communications, financial information, or legal theories into an AI platform. That is especially true when a lawsuit, government investigation, subpoena, employment dispute, contract dispute, or business conflict is already underway or reasonably expected.

A business owner facing a breach of contract claim, for instance, may be tempted to ask an AI platform to evaluate defenses, identify weaknesses, or draft a settlement strategy. But if the information entered into the platform is later discoverable, that business owner may have created evidence that can be used by the other side.

The safer course is to speak directly with counsel before using AI in connection with a legal dispute. If AI is going to be used at all, it should be used carefully, with an understanding of the platform’s privacy terms and only under appropriate legal guidance.

Conclusion

United States v. Heppner is an important warning about the risks of using public AI platforms for legal strategy or sensitive litigation analysis. The court made clear that communications with AI are not automatically protected by attorney-client privilege or the work-product doctrine. A public AI platform is not a lawyer, does not create an attorney-client relationship, and may not provide the confidentiality required for privilege.

For business owners, executives, professionals, and individuals involved in legal disputes, the lesson is simple: do not put sensitive legal information into a public AI platform and assume it will remain protected. When confidentiality matters, speak with an attorney first.