Estimated Reading Time: 6 minutes
- What Happened in Heppner
- Why the Attorney-Client Privilege Claim Failed
- Why the Work Product Claim Failed
- Why the Decision Matters
- A Note the Court Left Open
- Takeaways for Businesses
A federal court in New York has issued a warning that every business leader, in-house counsel, and HR professional should hear: sharing sensitive legal information with an AI chatbot may cost you your attorney-client privilege.
The ruling — one of the first of its kind — adds to a growing list of AI-related pitfalls in the legal system, including court filings that cite fabricated cases.
In U.S. v. Heppner, a federal district court considered whether a party’s interactions with Anthropic’s Claude qualified for protection under the attorney-client privilege. The court also considered the work product doctrine, which protects materials a lawyer prepares for a lawsuit. The court rejected both arguments. Prosecutors could review the defendant’s communications with Claude. Sensitive information can lose protection and become evidence the other side can obtain.
What Happened in Heppner
Bradley Heppner, an executive involved with several corporate entities, faced federal fraud charges. During the investigation, the FBI searched his home and seized documents and electronic devices. Among the materials were records of his exchanges with Anthropic’s Claude.
The court said those exchanges included reports and analyses that outlined possible defense strategy, including arguments tied to the facts and the law. Heppner argued the materials should be off-limits to prosecutors, protected by either attorney-client privilege or the work product doctrine.
The court rejected both arguments. Prosecutors got the communications.
Why the Attorney-Client Privilege Claim Failed
Attorney-client privilege depends on a few core elements. The client must communicate with counsel in confidence for the purpose of seeking or receiving legal advice. In Heppner, the court found those elements missing.
Claude Is Not a Lawyer
Heppner communicated with Claude, not with his attorneys. He argued he used Claude as a way to organize his thinking before speaking with actual counsel. The court wasn't persuaded. Discussing legal issues with a non-lawyer, even as preparation for a later conversation with an attorney, does not make those discussions privileged.
Business leaders should pay attention to that point. Executives, HR teams, founders, and in-house staff often use AI to organize facts before contacting counsel. Heppner suggests that work may fall outside the privilege when it happens on a third-party platform.
The Platform Itself Broke Confidentiality
Privilege requires confidentiality. Once a client shares protected information with a third party outside the attorney-client relationship, the client may waive that protection.
The court treated Claude as a third-party commercial service, not a protected legal channel. Anthropic's own policies made the problem worse: the company collected user prompts and Claude's responses, reserved the right to use that data to train its models, and could disclose it to third parties, including government authorities.
In the court's view, Heppner hadn't shared his legal analysis with his lawyer. He shared it with a commercial platform whose own policy reserved rights over the data he entered.
Sending the Files to His Lawyers Afterward Didn’t Help
Heppner later gave the Claude-generated materials to his lawyers. The court said that did not change their status.
Privilege usually depends on how a document came into existence and whether the client kept the communication confidential from the start. A client cannot create a document outside the attorney-client relationship, share it with a third party, and then make it privileged by sending it to counsel afterward.
Why the Work Product Claim Failed
Heppner also claimed protection under the work product doctrine. In plain terms, work product usually means notes, research, strategy documents, or similar materials that a lawyer prepares, or directs someone else to prepare, for litigation or an expected dispute.
The court found that counsel did not prepare the documents, did not direct their preparation, and did not embed defense strategy in them. Heppner created the materials on his own.
Employees and executives may think they help the legal process when they use AI. If they do that work without counsel’s direction and through a third-party platform, a court may treat the material as discoverable, which means the other side may obtain and review it during litigation.
Why the Decision Matters
Heppner marks the first reported decision to directly address whether a party’s interactions with generative AI can waive attorney-client privilege and work product protection. Other courts may refine the analysis, but this case sets an early marker.
Old habits now create risk. People who share legal theories, sensitive facts, or prior attorney communications with an AI platform may create material the other side can later demand in a dispute.
A Note the Court Left Open
The opinion wasn't entirely without nuance. The judge acknowledged that the outcome might have been different if Heppner's attorneys had specifically directed him to use Claude in a particular way. In that scenario, the AI tool might have functioned more like an agent assisting counsel, potentially bringing the communications closer to protected territory.
Because Heppner acted on his own initiative, the court didn't go there. But the door isn't completely closed for future cases.
Takeaways for Businesses
Heppner isn't an edge case. It reflects how many employees already work: using AI to analyze contracts, draft responses, summarize disputes, and test arguments before looping in legal or HR.
Here's what organizations should consider in light of the judgement:
- Do not assume AI interactions are confidential. Sharing legal strategy, sensitive facts, or attorney communications with a third-party AI platform may waive privilege.
- Privilege depends on the relationship, not the topic. A discussion about legal issues does not qualify for protection unless it occurs within a protected attorney-client relationship.
- Sending the material to counsel later may not fix the problem. If the material lacked privilege when the client created it, forwarding it to a lawyer does not automatically make it protected.
- Work product protection also has limits. Materials created through AI may fall outside that protection when counsel did not prepare them or direct their preparation.
- AI policies should address legal risk. Internal policies should make clear that employees should not enter legal strategy, threatened claims, internal investigations, or dispute-related facts into third-party AI tools without attorney guidance.
- Train your teams. Executives, managers, HR teams, and employees should understand that using AI to summarize disputes or test legal theories can create discovery risk.
- Get counsel involved early. When a matter might involve litigation, a government investigation, employment claims, or regulatory scrutiny, don't wait. The earlier an attorney is directing the work, the stronger the protection.
Convenience has always been the enemy of confidentiality. Heppner is a reminder that the rules of privilege don't bend for efficiency, and that the cost of a quick AI query can be far higher than it appears.
At Strategy Law, LLP, we advise businesses, executives, and individuals on legal issues involving technology, privacy, governance, and litigation. To discuss how these developments may affect your organization, contact us at (408) 478-4104 or reach out online to schedule a consultation.
This article is for general informational purposes only and does not constitute legal advice. Reading it does not create an attorney-client relationship. For advice about your specific situation, consult a licensed attorney.