Southern District of New York judge rules that AI-generated documents prepared without counsel direction were not protected by privilege

The issue arose in an ongoing criminal case charging the defendant with securities fraud, wire fraud, conspiracy to commit securities and wire fraud, making false statement to auditors, and falsification of records.

In the course of discovery, a privilege log provided by the defendant’s counsel identified thirty-one documents described as “[a]rtificial intelligence-generated analysis conveying facts to counsel for [the] purpose of obtaining legal advice[,]” generated by Claude, an AI tool created by third-party company, Anthropic.

Attorney-Client privilege generally “protects communications (1) between a client and his or her attorney (2) that are intended to be, and in fact were, kept confidential (3) for the purpose of obtaining or providing legal advice.”¹  The work-product doctrine “provides qualified protection for materials prepared by or at the behest of counsel in anticipation of litigation or for trial.”² However, its application to AI-generated documents is an issue courts are only now beginning to face. 

In this case, the government argued that the AI-generated documents were not subject to the attorney client privilege because (1) they were not communications between the defendant and his counsel (even if later sent to the attorney); (2) they could not have been created for the purpose of obtaining legal advice, even if later transmitted to counsel for that purpose, in part because Anthropic’s policies explicitly disclaims its use for the purpose of gaining legal advice; and (3) they were not confidential, because the defendant’s queries had been entered into an AI system whose policies explicitly disclaimed confidentiality and instead permitted disclosure to government authorities.³ The government further argued that the subsequent act of transmitting these AI-generated documents to counsel “does not create a shield of attorney-client privilege[.]”⁴ 

As to the attorney work product question, the government argued that no work product protection could apply because the defendant had taken it upon himself to generate these documents and not done so at the direction of his counsel (a fact the defense counsel conceded).⁵  Even if the defense counsel later used these documents (and created work product incorporating them), it would not retroactively create protection for the AI-generated documents themselves.

Judge Rakoff issued his ruling from the bench during a pre-trial conference held Tuesday, February 10, where he agreed with the government in finding that there was no basis for a claim of attorney-client privilege or work product. He further indicated that a written order will follow in due course.

Key takeaways

While the application of attorney-client privilege and work product protections to AI-generated documents will surely raise novel issues, this ruling appears to rely on traditional principles of law. 

Importantly, Judge Rakoff’s ruling does not mean that the use of AI tools to generate legal analysis can never lead to documents that would be protected by the attorney-client or work product privilege. The applicability of such protections to AI-generated documents is jurisdiction and fact specific and must be evaluated case-by-case. Accordingly, to the extent a company or individual seeks to utilize AI in connection with litigation or other legal analysis and without direct involvement of counsel, they should ensure that (1) they are familiar with the policies and parameters of any AI tools they use (for example, whether it is a closed-end system or an open-ended system), and (2) that they are careful about how they document their use of the AI – for example, where appropriate, including in any prompts the purpose of the query and the involvement of counsel. 

Working with AI generated evidence was one of nine key challenges identified in the 2026 A&O Shearman Cross-border white collar crime and investigations review, see AI and investigations: from a helping hand to a source of evidence.  The issue of privilege and AI generated evidence is jurisdiction specific.  The A&O Shearman disputes practice is monitoring global developments on this issue closely.  If you would like to discuss this issue and how it relates to your business please contact your normal A&O Shearman contact. 

Footnotes:

1. United States v. Mejia, 655 F.3d 126, 132 (2d Cir. 2011)) 
2. (In re Grand Jury Subpoenas Dated Mar. 19, 2002, and Aug. 2, 2002, 318 F.3d 379, 383 (2d Cir. 2003))
3. Motion for a Ruling that Documents the Defendant Generated Through an Artificial Intelligence Tool Are Not Privileged, United States v. Heppner, No. 25-cr-00503-JSR (S.D.N.Y. Feb. 6, 2026), Dkt. No. 22
4. Id. 
5. Id. at 11