Client Alert: When AI-Generated Documents Are Not Legally Protected

What Your Organization Needs to Know After United States v. Heppner

May 8, 2026  

As organizations are increasingly using generative AI, a recent federal court decision makes clear that AI is no substitute for a licensed attorney, and that materials generated through AI tools may not be legally protected from disclosure. In United States v. Heppner, the U.S. District Court for the Southern District of New York ruled that documents created by a criminal defendant (without the direction or oversight of his attorneys) using the AI tool, Claude, were not shielded by attorney-client privilege or the work-product doctrine.

The bottom line: individuals and organizations should not use AI tools for legal advice or legal research without the express direction and oversight of legal counsel.

The court found that the AI-generated documents lacked the essential elements of attorney-client privilege: (1) a communication between a client and an attorney, (2) that was intended to be and was kept confidential, and (3) made for the purpose of obtaining or providing legal advice.¹ The court identified three reasons why documents generated through Claude were not privileged:

I. Claude Is Not an Attorney

Claude explicitly disclaims that it provides legal advice and cannot be classified as an attorney. Attorney-client privilege protects only communications between a client and a licensed attorney. Because the defendant independently used an AI tool without any attorney involvement privilege did not apply.

II. The Communications Were Not Confidential

Anthropic’s privacy policy for its consumer version of Claude permits the company to collect data on user prompts and AI outputs, use that data to train its models, and disclose it to third parties, including government authorities. This means users of AI tools have no reasonable expectation of confidentiality, which is required for privilege to apply. Many AI tools have similar data-sharing terms.

III. The Defendant Did Not Use AI at the Direction of Counsel

Heppner used AI on his own initiative after retaining counsel, but his attorneys did not direct him to do so. Because the communications with Claude were not made at counsel’s direction or for the purpose of obtaining legal advice from an attorney, the court held the communications fell outside the scope of privilege.

The court also rejected the defendant’s work-product doctrine claim. The work-product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation. Because Heppner created the AI documents on his own and not at the direction of his attorneys, and because the materials did not reflect counsel’s mental processes or legal strategy, they were not protected as work product.

What This Means

Heppner makes clear that AI-generated materials created without attorney involvement are unlikely to be protected by privilege or work-product doctrine, regardless of whether they are later shared with lawyers. This includes materials generated on enterprise or other paid versions of AI tools that otherwise allow for greater privacy and training restrictions. Sending unprotected documents to your attorney after the fact does not make them privileged. This means that AI-generated materials may be discoverable in litigation, government investigations, and regulatory proceedings.

AI Materials May Be Subject to Discovery

Under Federal Rule of Civil Procedure 34, parties in litigation may request any relevant electronically stored information (ESI). In practice, an opposing party or the government could request:

  • AI Inputs: Any prompt or question entered into an AI tool relating to a legal issue or the subject matter of a dispute.
  • AI Outputs: Any response generated by the AI tool in reply to those prompts.
  • Usage Data: Information about how the AI tool was used, including account records and metadata.

Organizations should assume that any use of AI tools may be discoverable, particularly in the context of anticipated or pending litigation. Do not rely on privilege or confidentiality agreements to shield AI-generated content.

Recommended Steps for Organizations That Have Already Used AI for Legal Research

If your organization has already used AI tools to seek legal advice or generate legal materials, we recommend the following steps:

  • Review Platform Terms: Examine the terms of service and privacy policies of any AI tools your organization has used. If the platform reserves the right to collect, train on, or disclose user data, privilege is unlikely to apply to those interactions.
  • Disclose AI Use to Legal Counsel: Inform your attorneys about all uses of AI tools for legal research or analysis so they can assess any impact on privilege and litigation strategy.
  • Establish Internal AI Use Policies: Adopt clear policies restricting the use of AI tools for legal matters. Employees should use only approved, enterprise-grade AI tools for approved purposes, and should never input privileged or confidential information into consumer-grade AI tools. Further, employees should only use AI tools for legal matters at the express direction of, and with strict oversight by, internal or outside legal counsel.

¹United States v. Mejia, 655 F.3d 126, 132 (2d Cir. 2011); see also Upjohn Co. v. United States, 449 U.S. 383, 389 (1981).

For further information contact Cameron Robinson (crobinson@crokefairchild.com) or Mackenzie Custer (custerme@mail.uc.edu).