AI Tools, Legal Advice and the Limits of Attorney-Client Privilege

Summary — AI chat logs, including prompts and outputs generated through tools such as ChatGPT, may be discoverable in litigation when they are relevant, nonprivileged and proportional to the needs of the case. 

Recent court decisions indicate communications with public AI platforms are unlikely to be protected by attorney-client privilege when they are not confidential, are not made to an attorney or are not connected to legal advice. Accordingly, organizations using AI in connection with legal matters should avoid public tools for sensitive information and ensure any AI-generated content is tied to confidential communications with counsel.

Business use of Large Language Models (“LLMs”), like ChatGPT and Claude, is growing rapidly. Employees across departments—legal, finance, sales, and the C-Suite—now routinely use AI platforms to draft content, analyze data, and answer questions. 

As with email and other electronic communications, chat logs containing LLM prompts and outputs (“AI Chat Logs”) are discoverable in litigation when they fall within the scope of discovery. See, e.g., In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-MD-3143, 2025 WL 3468036, at *3-5 (S.D.N.Y. Dec. 2, 2025) (ordering the production of ChatGPT logs as they were relevant to and proportional to the needs of the case). Under the Federal Rules of Civil Procedure, AI Chat Logs are discoverable if they are nonprivileged and relevant to any party’s claim or defense and proportional to the needs of the case, considering the importance of the issues at stake in the action, the amount in controversy, the parties’ relative access to relevant information, the parties’ resources, the importance of the discovery in resolving the issues, and whether the burden or expense of the proposed discovery outweighs its likely benefit. Fed. R. Civ. P. 26(b).

What happens, then, when an employee uses an open LLM[1] for legal questions?  Are those prompts and outputs protected by the attorney-client privilege?

Based on the Heppner ruling, the answer may depend on the circumstances—and courts are signaling skepticism about attorney-client privilege claims for open AI platforms. In U.S. v. Heppner, Case No. 25 Cr. 503, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026), the court held that a defendant’s communications with an AI platform and the material generated from those communications were not protected from disclosure by the attorney-client privilege or the work product doctrine. The court rejected the privilege claim for four reasons:

• Claude is not an attorney, so communications with it are not attorney-client communications;
• The communications were not confidential—Anthropic’s privacy policy permits collecting and sharing user inputs and outputs;
• The defendant did not use Claude at counsel’s direction; and
• The outputs did not reflect defense counsel’s strategy or work product.
   

Id. While this is a criminal case and therefore does not fall under the Federal Rules of Civil Procedure, the ruling demonstrates that AI Chat Logs from an open AI source may not be protected from disclosure by the attorney-client privilege.

The analysis may differ when an organization uses a closed-source, enterprise LLM rather than a publicly available tool.[2] Court rulings analyzing whether attorney notes created before litigation is reasonably anticipated[3] may be instructive in determining whether the attorney-client privilege protects AI Chat Logs created by in-house counsel or employees before litigation is reasonably anticipated from disclosure in civil litigation.

Briefly, the attorney-client privilege “affords confidential communications between lawyer and client complete protection from disclosure.”  Hawkins v. Stables, 148 F.3d 379, 383 (4th Cir. 1998).  However, it is strictly construed and applies only if:

(1) the asserted holder of the privilege is or sought to become a client; (2) the person to whom the communication was made (a) is a member of the bar of a court, or his subordinate and (b) in connection with this communication is acting as a lawyer; (3) the communication relates to a fact of which the attorney was informed (a) by his client (b) without the presence of strangers (c) for the purpose of securing primarily either (i) an opinion on law or (ii) legal services or (iii) assistance in some legal proceeding, and not (d) for the purpose of committing a crime or tort; and (4) the privilege has been (a) claimed and (b) not waived by the client.

In re Grand Jury Subpoena, 341 F.3d 331, 335 (4th Cir. 2003) citing United States v. Jones, 696 F.2d 1069, 1072 (4th Cir. 1982).

Courts that have ruled that an attorney’s notes met the elements of the attorney-client privilege found that the notes reflected communications with either a client or in the case of in-house counsel, outside counsel. TransWeb, LLC v. 3M Innovative Props. Co., No. 10-CV-4413 FSH PS, 2012 WL 2878076, at *17–18 (D.N.J. Apr. 12, 2012), report and recommendation adopted as modified, No. CIV. 10-4413 FSH PS, 2012 WL 2878075 (D.N.J. July 13, 2012) (finding in-house counsel’s notes protected from disclosure by the attorney-client privilege because the notes referenced a conversation with an employee memorializing confidential communications and were drafted by in-house counsel in advance of a call with outside counsel and communicated to outside counsel during the call); McCook Metals L.L.C. v. Alcoa Inc., 192 F.R.D. 242, 255 (N.D. Ill. 2000) (holding an employee’s handwritten notes on a draft letter were protected from disclosure by the attorney-client privilege because they reflected the employee’s discussion with an attorney as to the wording in the letter).

In those cases where courts have held that attorney notes are not protected from disclosure by the attorney-client privilege, no evidence existed to demonstrate that the information in the notes was communicated to the client or a communication from the client to the attorney for the purpose of rendering legal advice.  Cencast Services, LP v. United States, 91 Fed.Cl. 496, 507 (Fed.Cl. 2010) (finding that the attorney-client privilege did not apply as the attorney's notes did not reveal client communications and instead recounted her or another attorney's views of the significance of certain issues they encountered); Am. Nat'l Bank & Trust Co. of Chicago v. AXA Client Solutions, No. 00 C 6786, 2002 WL 1058776, at *2 (N.D.Ill. Mar. 22, 2002) (“The handwritten notes merely reflect in-house counsel's own uncommunicated thoughts, and such recorded and uncommunicated thoughts fall outside the province of the attorney-client privilege.”). 

These rulings suggest that AI Chat Logs may be discoverable unless affirmatively linked to privileged attorney-client communications. In-house counsel should consider the following steps to protect privilege when employees use LLMs in connection with legal matters:

• Prohibit use of open or unapproved AI tools for privileged materials. Do not input attorney advice, work product, investigation materials, or case strategy into any public or unapproved AI platform. Information entered into—or generated by—such tools may be discoverable by adverse parties.
• Link AI outputs to specific legal advice. Ensure that prompts and outputs tie to a specific communication with counsel seeking or reflecting legal advice. Simply forwarding AI-generated content to counsel, without more, is unlikely to establish privilege.
• Document the privilege connection. Maintain records showing that specific prompts and outputs were shared with counsel for the purpose of legal advice, including the date and context. Storing AI Chat Logs in counsel’s files, without documented legal purpose, is unlikely to be sufficient.
   

As AI tools continue to evolve and become additional sources of electronically stored information in civil discovery, the ability to tie prompts and outputs to specific confidential communications between counsel and client and to legal advice from counsel may assist in protecting specific AI Chat Logs containing privileged information from disclosure.