On February 17, 2026, Judge Jed Rakoff of the Southern District of New York issued a significant ruling in United States v. Bradley Heppner. The opinion squarely addressed whether conversations with a publicly available generative AI system can be protected by attorney-client privilege or the work product doctrine.

The court’s answer: not under the circumstances presented.

This decision has immediate implications for executives, compliance teams, and individuals who use public AI platforms to think through legal strategy.

𝐓𝐡𝐞 𝐅𝐚𝐜𝐭𝐬 𝐁𝐞𝐡𝐢𝐧𝐝 𝐭𝐡𝐞 𝐑𝐮𝐥𝐢𝐧𝐠

The case arose in a federal securities fraud prosecution. After learning he was under investigation, the defendant used a widely available AI chatbot to evaluate potential defenses, analyze legal exposure, and explore anticipated arguments from the government.

When authorities executed a search warrant in late 2025, they recovered dozens of documents reflecting those AI exchanges.

The defendant argued the materials were protected because:

• He had incorporated information obtained from counsel,

• He prepared the materials to assist in discussions with his attorneys, and

• He later shared the outputs with them.

However, defense counsel acknowledged they had not instructed him to use the AI tool.

𝐓𝐡𝐞 𝐂𝐨𝐮𝐫𝐭’𝐬 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬

Judge Rakoff applied traditional privilege doctrine and found multiple deficiencies.

1️⃣ No Attorney-Client Relationship

Attorney-client privilege protects communications between a client and a licensed attorney (or an agent acting on the lawyer’s behalf). A public AI system is neither. The court emphasized that privilege depends on a professional human relationship grounded in fiduciary duties and regulatory oversight. An algorithm cannot satisfy that requirement.

2️⃣ No Reasonable Expectation of Confidentiality

The AI provider’s published terms disclosed that user inputs and outputs could be collected, used for system improvement, and disclosed under certain circumstances. By agreeing to those terms, the defendant undermined any assertion that the communications were confidential.

Even if privileged information had originally been involved, sharing it with a third-party platform constituted waiver.

3️⃣ Not Communications for Legal Advice from Counsel

Although the defendant maintained that he intended to discuss the AI output with his lawyers, he independently chose to use the system. Because counsel did not direct or supervise the AI use, the chatbot was not treated as the lawyer’s agent.

The relevant inquiry became whether he was seeking legal advice from the AI — which does not trigger privilege protection.

𝐖𝐨𝐫𝐤 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐃𝐨𝐜𝐭𝐫𝐢𝐧𝐞

The court also rejected work product protection. That doctrine applies to materials prepared by or at the direction of counsel in anticipation of litigation. Since the defendant used the AI tool on his own initiative, and not under attorney instruction, the protection did not attach.

𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞 𝐈𝐦𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬

The lesson from United States v. Heppner is practical and immediate:

• Public AI tools are not confidential advisors.

• Terms of service can defeat privilege arguments.

• Independent AI use may waive otherwise protected information.

• Counsel involvement is critical if AI is used in litigation strategy.

For companies developing AI governance programs, privilege risk must now be explicitly addressed in policies, training, and vendor contracting.

This is no longer a theoretical concern. It is judicial precedent.