Federal Judge Rules AI Queries Lack Attorney-Client Privilege Protection

A federal judge in the U.S. District Court for the Southern District of New York, Judge Jed S. Rakoff, issued a significant ruling on February 10, 2026, determining that documents generated through interactions with a third-party AI tool are not protected by either attorney-client privilege or the work product doctrine. This decision, in the case of U.S. v. Heppner, is noted as potentially the first of its kind to address the privileged status of information shared with publicly accessible AI tools. The defendant, Bradley Heppner, had used Anthropic's Claude to generate 31 documents related to a fraud investigation, prior to his arrest and without direction from his counsel. Judge Rakoff's reasoning for denying attorney-client privilege centered on the diminished expectation of confidentiality when information is disclosed to a public AI tool. The court highlighted several factors: the communication occurred with a third-party AI platform, the AI tool's privacy policy explicitly stated collection of both user inputs and outputs, this collected data was used for training the AI, and the company reserved the right to disclose such data to various third parties, including government regulatory authorities. The work product doctrine was also deemed inapplicable because the defendant had not prepared the materials at the direction of counsel. This ruling underscores the critical importance for individuals and organizations to meticulously review the terms of service and privacy policies of any AI tool they utilize, particularly concerning data usage for training and disclosure rights. While the ruling is described as 'highly fact-specific,' its implications are broad, potentially reshaping how legal professionals and the public approach the use of AI for sensitive or confidential tasks, and prompting AI developers to consider more robust privacy safeguards for their services. This case sets a crucial precedent for the intersection of AI, law, and data privacy.

EU AI Act Art. 50 Compliant: This analysis is based solely on the provided source material. No external data or prior knowledge was used.