The rapid integration of generative artificial intelligence into legal practice has raised increasingly complex questions concerning confidentiality, legal professional privilege, and the treatment of AI-generated communications within litigation. Large language models are now routinely used to summarise evidence, organise chronologies, analyse contracts, and assist with legal drafting. Clients themselves also increasingly engage with artificial intelligence platforms before formally instructing legal advisers, often using public systems to evaluate legal exposure or test litigation arguments.

Against this background, the 2026 decision in United States v. Heppner has emerged as one of the first significant judicial authorities to consider whether communications involving a generative artificial intelligence platform may attract attorney-client privilege or work product doctrine protection. The decision is important not simply because it concerns artificial intelligence, but because it demonstrates how courts may apply conventional privilege principles to emerging technologies.

Rather than developing a new doctrinal framework for AI systems, the court approached the issue through established principles of confidentiality and waiver. In doing so, the judgment provides an important indication of how future courts may address disputes involving artificial intelligence and legally privileged material.

Background to the proceedings

The proceedings arose from a criminal investigation concerning allegations of securities fraud and wire fraud against Bradley Heppner. During the investigation, federal prosecutors obtained electronic materials that included 31 documents generated through the defendant’s use of Claude, Anthropic’s generative AI platform. The materials reportedly included analyses of legal risks, litigation strategy, and potential defence arguments.

The defendant argued that these documents were protected by attorney-client privilege and the work product doctrine because they reflected legal analysis prepared in anticipation of litigation. The U.S. Government challenged that position and sought disclosure of the materials.

The central question before the United States District Court for the Southern District of New York was whether communications generated through interaction with a public artificial intelligence platform could satisfy the legal requirements necessary for privilege protection.

On 17 February 2026, the court concluded that they could not.

Attorney-client privilege and confidentiality

The court analysed the dispute through orthodox principles governing attorney-client privilege. Under United States law, privilege generally protects confidential communications between a client and lawyer made for the purpose of obtaining or providing legal advice.

Judge Jed Rakoff held that the defendant’s interactions with the AI platform failed to satisfy these requirements.

First, the communications were not exchanges with a legal adviser. Claude was not a lawyer, could not provide legal representation, and could not enter into an attorney-client relationship. The court therefore treated the platform as an external third party rather than as an extension of legal counsel.

Second, and more significantly, the court found that the communications lacked the confidentiality required for privilege protection. The judgment relied heavily upon the AI developer’s published terms and privacy policies, which permitted the retention and review of user inputs and outputs and reserved rights to disclose information in certain circumstances.

In the court’s assessment, a user who voluntarily submitted information to such a platform could not reasonably assert that the communications remained confidential in the legal sense required for privilege.

The defendant further argued that the documents later formed part of communications with legal advisers. The court rejected this submission, holding that once confidential information had been voluntarily disclosed to an external third party, privilege could not subsequently be restored.

The judgment therefore concluded that the documents fell outside the scope of attorney-client privilege.

Work product doctrine and AI-generated materials

The defendant also argued that the materials were protected under the work product doctrine because they had been created in anticipation of litigation.

The court again rejected the claim.

Judge Rakoff distinguished between materials prepared by lawyers, or at their direction, and materials independently generated by a client using a public artificial intelligence system. The evidence suggested that the defendant’s use of the AI platform had not been directed or supervised by legal counsel.

As a result, the court concluded that the materials did not sufficiently reflect attorney strategy or legal mental impressions so as to attract work product doctrine protection.

Although Heppner arose within a criminal law context, the implications of the judgment extend well beyond criminal proceedings. The decision is likely to influence future disputes involving internal investigations, regulatory proceedings, civil litigation, and arbitration.

Public AI platforms as third-party intermediaries

One of the most significant aspects of Heppner is the court’s treatment of public artificial intelligence systems as third-party intermediaries.

Privilege doctrine has long depended upon the preservation of confidentiality. Communications disclosed to outsiders generally fall outside privilege unless the disclosure is necessary for the provision of legal advice. Courts have historically recognised limited exceptions involving interpreters, consultants, and expert witnesses engaged to assist legal counsel.

Heppner suggests that publicly accessible AI platforms may not fit comfortably within those recognised categories.

This distinction is likely to become increasingly important as businesses and law firms continue integrating artificial intelligence into legal workflows. Many public AI systems operate pursuant to terms permitting the retention, processing or review of user data. Such provisions may undermine any reasonable expectation of confidentiality.

By contrast, enterprise artificial intelligence systems designed specifically for legal or corporate use increasingly include contractual confidentiality protections, restricted data retention policies, and deployment within secure infrastructure controlled by the organisation itself.

Future courts may therefore distinguish between public consumer platforms and enterprise systems operating within professionally supervised environments.

Comparative observations under English law

Although United States v Heppner was decided under United States law, similar issues are likely to arise under English legal principles.

English law also treats confidentiality as fundamental to legal professional privilege. Legal advice privilege protects confidential communications between lawyer and client for the purpose of giving or obtaining legal advice, while litigation privilege protects certain confidential communications connected to adversarial proceedings. As under United States law, voluntary disclosure of privileged information to third parties may result in waiver.

Accordingly, the reasoning in Heppner may prove influential in future English disputes concerning generative artificial intelligence platforms. A court may conclude that uploading legally sensitive material into a public AI system constitutes disclosure to an external third party inconsistent with the maintenance of privilege.

Implications for legal practice

The practical consequences of Heppner are considerable. Law firms are increasingly likely to adopt stricter internal policies governing the use of artificial intelligence systems in legal work. Clients may also require express warnings against uploading privileged or commercially sensitive material into public AI platforms lacking appropriate confidentiality protections.

Corporate investigations and disclosure exercises may become more complicated as parties seek information concerning whether artificial intelligence systems were used during the preparation of legal analyses or internal reports. Opponents in litigation may increasingly argue that privilege has been waived through the use of public AI platforms.

English courts and regulators are likely to regard such arguments as instructive as equivalent disputes begin to arise before domestic tribunals. Solicitors regulated by the Solicitors Regulation Authority remain under continuing duties to safeguard confidential client information. The use of external artificial intelligence systems capable of retaining or processing sensitive legal material may therefore raise regulatory and professional conduct concerns in addition to privilege issues. 

Sources

https://harvardlawreview.org/blog/2026/03/united-states-v-heppner/?utm_source

https://www.paulweiss.com/insights/client-memos/sdny-court-considers-whether-ai-generated-documents-are-subject-to-privilege-protections?utm_source

https://www.lawfaremedia.org/article/ai-and-privilege-after-united-states-v.heppner?utm_source