AI-generated documents may not be protected by attorney-client privilege

The 10 February 2026 ruling by Judge Jed Rakoff of the US District Court for the Southern District of New York signals a significant and largely uncharted risk for clients who turn to AI tools while navigating legal proceedings. Read the full decision here.

The case centres on Bradley Heppner, who was arrested in November 2025 on charges of securities and wire fraud. Federal agents searching his home seized electronic devices containing approximately 31 documents that Heppner had generated using Anthropic’s AI tool Claude. After retaining legal counsel and receiving a grand jury subpoena, Heppner had used Claude on his own initiative to prepare reports outlining his defence strategy and potential legal arguments. He later transmitted those reports to his attorneys.

Heppner’s defence team argued the documents were privileged, describing them as a means of consolidating the client’s thoughts for the purpose of communicating with counsel. The government pushed back, contending that sharing information with a commercial AI tool, which operates under terms of service explicitly disclaiming user confidentiality, constitutes a disclosure to a third party, and therefore destroys any claim of privilege.

Judge Rakoff agreed with the government, holding that attorney-client privilege did not apply because Heppner had shared the communications with an AI tool, not a lawyer, and the AI tool did not maintain confidentiality. The applicable Anthropic privacy policy notes that user prompts could be used to train its model and might be disclosed to government authorities and third parties, a provision the court found fatal to the privilege claim.

The work product doctrine fared no better. That doctrine protects materials prepared by or at the direction of legal counsel in anticipation of litigation. Because Heppner acted on his own initiative, not at his lawyers’ direction, and because neither Heppner nor the AI tool are attorneys, Judge Rakoff declined to extend that protection as well.

The decision has immediate practical implications for anyone using AI tools in connection with legal matters. Enterprise-grade AI tools, which contractually commit to not training on user inputs and to maintaining input confidentiality, may be viewed differently by courts. Time will tell. For now, clients should avoid using AI tools for legal matters without clearance from legal counsel. Non-lawyers assisting counsel should make clear in their AI prompts that they are acting at a lawyer’s direction, and privilege logs should explicitly identify the AI tool used and the basis for any confidentiality expectation.

Now is a key time for law firms and their clients to revisit their AI usage policies, particularly the choice between consumer and enterprise AI platforms, before the next high-stakes matter lands in court.

---

Kathryn Nadro is a Partner in Levenfeld Pearlstein's Corporate Group, advising clients on cybersecurity, data privacy, and artificial intelligence matters. She is a Certified Information Privacy Professional (CIPP/US) and Artificial Intelligence Governance Professional (AIGP). 

Lisa Vandesteeg is a partner in and leader of LP’s Litigation Practice Group, and a member of the Financial Services & Restructuring Group. Her practice focuses on complex commercial litigation, bankruptcy, creditors’ rights, restructuring-related disputes, data security, and privacy.