DOJ: Conversations with Claude AI not legally privileged

The U.S. Department of Justice has recently made a significant legal argument regarding the interaction between a criminal defendant and Anthropic’s AI chatbot named Claude in a securities fraud case. According to the latest court filing in the United States v. Heppner case, the DOJ contends that conversations between the defendant, Bradley Heppner, and Claude do not fall under attorney-client privilege protections.

Heppner, who is facing charges related to securities fraud in the Southern District of New York, utilized Claude to generate around 31 documents seeking information about his legal situation before his arrest on November 4, 2025. Subsequently, federal authorities seized Heppner’s electronic devices, which contained these documents originating from his interactions with Claude.

In response to the defense counsel’s claim that these documents were protected communications, the prosecution presented a motion requesting access to the AI-generated files for potential use as evidence during the trial. The government’s argument against the privilege protection is threefold. Firstly, Claude is an AI tool and not a licensed attorney involved in the creation of the documents. Secondly, Anthropic’s terms of service explicitly state that legal advice is not within the scope of services provided. Thirdly, the defendant opted to share queries with a third-party platform, Claude, which has privacy policies allowing data collection, human review, and disclosure to authorities.

The DOJ emphasized that the defendant engaged with an AI tool for legal and factual information, not with actual legal representation, potentially undermining the claim of attorney-client privilege. The motion clarified that sharing non-privileged documents with counsel post-creation does not retroactively establish privilege protection. This case raises crucial questions about the confidentiality expectations when utilizing AI tools provided by commercial platforms like Anthropic.

Anthropic’s privacy policy, active during Heppner’s interactions with Claude, disclosed the collection of user inputs and AI outputs for model training purposes, specifying possible disclosure to government entities and other parties. Despite the defense counsel’s argument that Heppner generated the AI documents to seek legal advice from his attorneys, it was clarified that the attorneys did not direct the defendant to conduct searches using Claude.

Overall, this legal dispute sheds light on the intersection of AI technologies with legal frameworks, highlighting the nuances and complexities of establishing attorney-client privilege in the context of AI-driven interactions. It underscores the importance of understanding the limitations and implications of using AI tools for legal purposes within the framework of existing legal principles and regulations.