Federal Judge Rules That Use of AI Is Not Privileged

 

A dramatic change in confidentiality may be coming in the Courts.

Betsy Combier

betsy@advocatz.com

Editor, ADVOCATZ.com

Editor, ADVOCATZ Blog

Editor, NYC Rubber Room Reporter
Editor, Parentadvocates.org
Editor, New York Court Corruption
Editor, National Public Voice
Editor, NYC Public Voice
Editor, Inside 3020-a Teacher Trials

SDNY Judge Rules Defendant’s Own Use of AI Tools for Strategizing Not Privileged

By Kit Yona, M.A. | Reviewed by Joseph Fawbush, Esq. | Last updated on February 23, 2026

Attorney-client privilege exists to protect the relationship between a litigant and their legal counsel. Whether it’s a legal strategy, privileged materials, or legal advice shared between a defense counsel and their client, attorney-client privilege grants a reasonable expectation of confidentiality from the government compelling the sharing of privileged communications. Just make sure the information you share is with an actual attorney who is also a human being.

As evidenced by a memorandum released on February 17, 2026, by Southern District of New York (SDNY) District Court Judge Jed Rakoff, AI-generated documents do not fall under the attorney-client privilege, at least in this instance. Judge Rakoff denied the privilege claims asserted by Bradley Heppner, who had used Anthropic’s Claude generative artificial intelligence (AI) platform to devise a legal defense plan in anticipation of litigation after being indicted by a grand jury in November 2025 on charges including fraud, conspiracy, and falsifying corporate records.

A key takeaway in the court’s ruling against protecting Heppner’s AI-generated materials is that he did so of his own volition and not as an instruction from his attorney. This undermined his argument for including the AI documents seized by federal agents under both the attorney-client privilege and the work product doctrine. The ruling sets yet another precedent in the interactions between the legal system and generative AI platforms.

Privilege Requires a Trusting Human Relationship

Defense attorneys and their clients need to be able to discuss sensitive aspects of the case without fear that prosecutors will subpoena any privileged information shared. This is the basis behind the attorney-client privilege, which protects the following from mandatory disclosure:

• Communications between a client and their attorney
• Communications intended to be confidential and kept so
• Communications for the purpose of obtaining or providing legal advice

Attorney-client privilege is often paired with work-product protection, which is an offshoot of the relationship. The work-product doctrine is designed to shelter the attorney’s mental processes as they examine, analyze, and prepare their client’s defense.

Heppner wasn’t exactly in desperate need of help with his legal matters. Heppner, the founder of the financial services startup Beneficient, was indicted by a grand jury for allegedly looting more than $150 million from GWG Holdings, Inc., where he served as chairman. Charged with securities fraud, wire fraud, conspiring to commit wire and securities fraud, giving false statements to auditors, and falsifying corporate records, Heppner hired counsel from Quinn Emanuel for his defense. However, he decided to do a bit of research on his own as well.

A grand jury subpoena signaled that Heppner was almost certainly facing arrest soon. This prompted him to access the public AI tool Claude. Heppner had “conversations” with the generative AI tool in which he requested the creation of a defense strategy against the likely charges he’d be facing and suggestions on how to argue against the facts presented. Seizures made as part of a search warrant of Heppner’s home uncovered electronic devices containing Claude’s responses.

Heppner filed to have the results ruled protected work products and also covered by the attorney-client privilege. The government agreed to a “Privilege Protocol Stipulation” that kept the AI documents in question segregated and not examined until the matter was resolved.

Not on the Advice of Counsel

Heppner’s motion argued that his input into Claude included information he’d received from his counsel. He also stated that he planned to share the documents with his defense attorneys and to use the results to request case-specific advice. This, he claimed, made his use of generative AI and the results yielded covered by the attorney-client privilege.

Judge Rakoff’s memorandum from February 17 expanded on the court’s decision to orally grant the federal government’s request a week earlier, which made the results from Claude exempt from either the attorney-client privilege or the work product doctrine. Claude is not a lawyer and will issue a disclaimer when asked for legal advice. Because Anthropic’s terms of service (TOS) clearly state that all user input and AI prompts may be used to further train Claude, any interactions with the platform have no expectation of confidentiality or privacy. The TOS also notes that Anthropic has the right to share any user input with authorities.

The largest issue that Heppner’s action was unable to overcome was the fact that his actual defense counsel did not instruct him to use Claude to craft a legal defense. Since his living, breathing attorneys hadn’t advised him to share his user data with an AI model, the output from Claude had no attorney-client privilege. This also torpedoed any work product protections, as it was not produced by his actual legal team.

Given the alleged evidence the government claims to have against him, being thwarted in an attempt to suppress his AI prompts may not be the deciding issue. Still, having a jury hear about his chat with Claude isn’t likely to help.

Related Resources

• Should Attorneys Ever Cite to AI? If So, How? (FindLaw’s Practice of Law)
• Lindell Case Indicates Artificial Intelligence and the Legal System Are Still a Work in Process (Findlaw’s Practice of Law)
• Why Judges Are Getting Creative With AI-Abusing Attorneys (FindLaw’s Practice of Law)