Client’s Use of AI Not Protected by Attorney Client Privilege

A Case of First Impression - What Does It Mean for Your Use of AI Tools?

On February 17, 2026, Judge Rakoff of the Southern District of New York held that an AI-chat made by a criminal defendant in anticipation of a meeting with his attorneys was not protected by either attorney-client privilege or the work product doctrine. This ruling—which the court acknowledged answers “a question of first impression nationwide”—has serious implications for anyone using AI tools in the context of legal matters. Every business and legal team should consider AI usage through the lens of whether privilege attaches and whether waiver may occur.

Background

Bradley Heppner, a former executive of GWG Holdings, Inc., was indicted on October 28, 2025, on five federal felony counts involving alleged fraudulent activity. At the time of Heppner’s arrest, FBI agents seized numerous documents and electronic devices from Heppner's home, including approximately thirty-one communications between Heppner and the AI tool Claude. Heppner’s counsel claimed privilege over the documents and the Government moved for the exchanges with Claude (the “AI Documents”) to be disclosed.

Attorney-Client Privilege

The court found that the 31 AI Documents lacked at least two, if not all three, elements required for attorney-client privilege: (1) a communication between a client and an attorney that was (2) intended and kept confidential, and (3) made for the purpose of obtaining legal advice.

• No Attorney-Client Relationship. Because Claude is not an attorney, the fundamental predicate for the privilege is absent. In rejecting arguments comparing AI inputs to word processing software use, the order reasoned that all recognized privileges require “a trusting human relationship,” such as one with “a licensed professional who owes fiduciary duties and is subject to discipline.”
• No Confidentiality Due to Claude's Terms and Conditions. Relying on Anthropic’s terms and conditions, the decision noted that Anthropic collects data on users' inputs and uses the data to train the AI model underlying Claude. The terms specifically reserve the right to disclose user data to "third parties," including "governmental regulatory authorities." Accordingly, Heppner had no reasonable expectation of confidentiality.
• No Legal Advice Sought from Claude. The court also found that Heppner did not communicate with Claude for the purpose of obtaining legal advice. Although Heppner’s counsel argued that Heppner created the AI Documents for the "express purpose of talking to counsel," counsel admitted that Heppner "did not do so at the suggestion or direction of counsel." Because Heppner "communicated with Claude of his own volition," the court focused on whether he intended to obtain legal advice from Claude—not whether he later shared the outputs with counsel. When asked, Claude itself disclaims providing legal advice, having told the Government when asked: "I'm not a lawyer and can't provide formal legal advice or recommendations.”

Work Product Doctrine

Similarly, the work-product argument was rejected on two grounds: the AI Documents were not prepared "by or at the behest of counsel," and they did not reflect counsel's mental processes or strategy.

• Not Prepared at the Direction of Counsel. Under Second Circuit law, the work‑product doctrine applies only to work performed by an attorney or the attorney’s agent. Applying that standard, the analysis focused on whether Heppner was acting as counsel’s agent when he communicated with Claude. Because he was not, the AI Documents fell outside work-product privilege.
• Did Not Reflect Counsel's Strategy. Although the AI Documents may have "affect[ed]" counsel's strategy going forward, they did not "reflect" counsel's strategy at the time Heppner created them. Therefore, without involvement of counsel in creating the materials, the policy rationale for protection does not apply. 

Key Takeaways

Consider the confidentiality of the AI tools you use. Clients should avoid discussing legal or factual issues with AI tools that do not provide for the confidentiality of uploaded data, prompts, or responses. Consider policies that require employees to use approved internal AI tools and prohibit the use of unsanctioned platforms.

Discuss AI use with counsel before using AI in connection with potential or active litigation. Companies and individuals facing potential litigation should consult with legal counsel before using AI tools—even internal, private tools. Involve the legal department or outside counsel as early as possible to ensure the greatest protection. Seek guidance from counsel on proposed uses of AI tools.

Legal teams should proactively communicate this risk to business stakeholders. Members of the business team need to understand the dangers of turning to AI before engaging the legal team, so they do not inadvertently waive privilege or create discoverable documents.

Womble Bond Dickinson (US) LLP’s White Collar Defense and Criminal Investigations Team navigates domestic and international clients in all manner of white collar, regulatory, corporate and congressional investigations. Our team includes a distinguished roster of veteran defense attorneys, former federal prosecutors and U.S. Attorneys who served at the highest levels of the Department of Justice and at leading United States Attorneys’ Offices. Our team includes Chambers Ranked (Band 1) lawyers and alumni of the U.S. Department of Justice, the SEC’s Enforcement Division, the U.S. Senate, House of Representatives, and in-house compliance specialists of publicly traded companies.