SDNY Rules AI-Generated Legal Strategy Lacks Attorney-Client Privilege

The intersection of generative AI and the legal system has reached a critical turning point following a precedent-setting ruling in the Southern District of New York. In the case of USA vs. Heppner, U.S. District Judge Jed S. Rakoff addressed a 'question of first impression' that carries profound implications for how defendants, corporations, and legal counsel interact with artificial intelligence. The court determined that approximately 31 documents generated by defendant Bradley Heppner using a publicly available AI platform—likely Claude, given the entity profile—did not qualify for attorney-client privilege or work product protection. This decision underscores a fundamental misunderstanding among lay users regarding the 'confidential' nature of AI interactions and sets a high bar for the technical environments required to maintain legal secrecy.

The core of the dispute arose after Heppner, facing charges of securities and wire fraud, independently used a generative AI tool to draft defense strategies, analyze the government’s likely case theory, and explore his potential criminal exposure. Crucially, Heppner conducted these activities without the direction or knowledge of his legal team, only sharing the results with them after the fact. When the FBI executed a search warrant on Heppner’s property in November 2025, they seized electronic devices containing not only the AI-generated summaries but also the raw interaction logs and prompts. The defense’s attempt to claw these materials back under privilege was rejected because the defendant had effectively 'published' his thoughts to a third-party service provider, thereby waiving the confidentiality essential to the privilege.

“

The court determined that approximately 31 documents generated by defendant Bradley Heppner using a publicly available AI platform—likely Claude, given the entity profile—did not qualify for attorney-client privilege or work product protection.

From a legal standpoint, the ruling hinges on the failure to meet the requirements of the work product doctrine. For a document to be protected as work product, it must be prepared in anticipation of litigation by or at the direction of an attorney. Because Heppner acted on his own initiative, the court viewed his AI interactions as personal reflections rather than professional legal preparation. Furthermore, the use of a consumer-grade, public AI platform meant that the data was subject to the platform's terms of service, which often include provisions for data retention and model training. This technical reality destroys the 'reasonable expectation of privacy' that is the bedrock of attorney-client communications.