E-discovery Case Law Alerts

AI Conversations with Public Chatbots Are Not Protected by Legal Privilege

In a landmark ruling on the outputs of LLM applications (in this case, Anthropic's Claude), Hon. Jed Rakoff of the Southern District of New York ruled that AI outputs, even on legal topics, are not privileged.

Download PDF Version

United States v. Heppner, No. 25-cr-00503-JSR (S.D.N.Y. Feb. 10, 2026)

Why This Alert Is Important

This landmark ruling confirms that traditional privilege doctrines apply strictly to new technologies, leaving no room for "AI-client privilege." It serves as a stark warning for corporate legal departments that employee use of consumer-grade generative AI for legal research or strategy can create a discoverable trail of evidence that is beyond the protection of the attorney-client privilege or the work product doctrine.

Overview

In United States v. Heppner, federal prosecutors sought access to 31 documents generated by Bradley Heppner, a financial executive accused of orchestrating a $150 million fraud scheme. After realizing he was the target of a law enforcement investigation, Heppner used Anthropic’s AI tool, Claude, to analyze his legal exposure, inputting prompts about the government’s investigation and potential defenses. He later shared these AI-generated analyses with his defense counsel at Quinn Emanuel.

When federal agents seized Heppner’s electronic devices, his attorneys asserted privilege, arguing the documents contained "AI-generated analysis conveying facts to counsel for the purpose of obtaining legal advice." The defense contended that because the documents incorporated legal information Heppner received from his lawyers and were intended to aid his defense, they should be protected. The government moved to compel, arguing that an AI tool is not a lawyer and that the public nature of the platform waived any expectation of confidentiality.

Judge Jed S. Rakoff of the Southern District of New York issued an oral ruling from the bench in favor of the government, finding no basis to shield the documents from discovery.

Ruling Summary

AI Is Not a Privileged Party
The court held that the attorney-client privilege did not apply because the communications were not between a client and a licensed attorney. Since an AI tool plainly is not an attorney, holds no law license, and owes no professional duties of loyalty or confidentiality to the user, it does not merit being treated as “privileged.” Because the communications involved a third-party platform rather than a legal professional, they failed the fundamental "privileged parties" test required for protection.
Third-Party Disclosure Waives Confidentiality
The ruling emphasized that the AI platform’s terms of service explicitly disclaimed confidentiality, stating that user inputs are not private and may be used for training. By voluntarily sharing case details with a commercial AI tool, Heppner disregarded any "reasonable expectation of privacy." The court reasoned that using a public chatbot is legally equivalent to "discussing the case with a friend" or conducting research in a public library, neither of which creates privilege.
No Retroactive Protection for Independent Work
Judge Rakoff rejected the argument that sending the AI documents to counsel after the fact could retroactively cloak them with privilege. Furthermore, the work product doctrine did not apply, because Heppner created the documents on his own initiative rather than at the direction of counsel. The court held that the doctrine protects the legal strategy of an attorney, not the independent internet research or autonomous musings of a layperson using an AI tool.

A lot... A LOT... is being written about this case in various forums. This might be one of the most watched and discussed tech-related cases in a decade. So, what are the takeaways from Judge Rakoff's opinion? First, it states that all recognized privileges require "a trusting human relationship" with someone who owes fiduciary duties.

We already know that Claude and similar products are not lawyers. Ask them and they'll tell you so. They do not have a license to practice law, therefore no privilege can be attached to their output. Additionally, the Terms of Service of these applications state inputs are not confidential , meaning hitting "enter" waives the right to privacy, and in turn, the expectation that privilege is similarly relinquished. And finally, sending unprivileged AI documents to your lawyer after said documents have been generated doesn't retroactively establish privilege.

I'm looking forward to the next case and the next judge that is likely to offer a different opinion and ruling. This topic is long from settled.

Nancy Patton, Esq., CEDS, Senior Director, Solutions Engineering, Exterro

Data Privacy Tip

For corporate legal teams, this ruling underscores the urgent need to establish clear AI-governance policies. To preserve privilege, legal research and analysis involving AI must be conducted within "closed" enterprise environments that guarantee confidentiality and must be performed under the direct supervision of counsel. Organizations should audit employee use of consumer AI tools to ensure that sensitive legal inquiries are not being inadvertently transformed into discoverable evidence.