Judge: AI Chat Logs Not Protected in Fraud Case
▼ Summary
– A US federal judge ruled that a defendant’s AI chatbot conversations about his legal case are not protected by attorney-client privilege, as the AI is not a lawyer.
– The judge also found no work-product protection because the defendant used the public AI platform without his attorney’s direction.
– The ruling determined there was no reasonable expectation of confidentiality, as the AI’s terms of service allow data collection and disclosure.
– In contrast, other recent court cases have protected AI conversations as work product for self-represented litigants under different procedural rules.
– Major law firms are now advising clients to avoid public AI for legal matters and to use only private, confidential systems under attorney guidance.
A recent federal court decision has sent a clear warning to anyone involved in legal proceedings: conversations with public AI chatbots are not confidential. In a first-of-its-kind ruling, a judge determined that these exchanges lack the legal protections traditionally afforded to attorney-client communications, meaning they can be obtained and used as evidence. This precedent has triggered a swift response from the legal industry, with major firms urgently advising clients on the significant risks of using generative AI for legal strategy.
The case, United States v. Heppner, centered on defendant Bradley Heppner, the former chairman of GWG Holdings. Facing securities and wire fraud charges in late 2025, Heppner consulted the AI assistant Claude before formally retaining defense counsel. He used the platform to analyze his potential legal exposure and explore defense arguments. When federal investigators later seized approximately 31 documents summarizing these AI interactions, Heppner claimed they were protected by attorney-client privilege and the work-product doctrine.
Judge Jed Rakoff of the Southern District of New York rejected these claims in a February 2026 ruling. His decision rested on three key points. First, privilege only exists between a client and a licensed attorney. An AI system is not a lawyer, holds no law license, and owes no duty of loyalty or confidentiality. As the judge noted, disclosing information to an AI is effectively disclosing it to a third party without any obligation of secrecy.
Second, Rakoff found there could be no reasonable expectation of privacy. He reviewed Anthropic’s terms of service and privacy policy, which explicitly state that user inputs and outputs may be collected, used to train models, and disclosed to third parties, including government authorities. By accepting these terms, Heppner consented to a framework of data use fundamentally at odds with legal confidentiality.
Third, the work-product protection did not apply because Heppner was not acting under the direction of his attorneys when he used Claude. The documents reflected his own independent inquiries, not the strategic legal thinking of his counsel.
This ruling appears to contrast with decisions in two other federal cases where judges protected AI conversations under the work-product doctrine. However, legal analysts highlight crucial factual distinctions. In those civil cases, the individuals were representing themselves without lawyers, and specific procedural rules offered broader protection. Heppner, a represented criminal defendant who consulted AI without attorney guidance, fell outside those safeguards. The courts in all three cases were careful to note they were not establishing universal rules for every situation.
The practical consequences have been immediate and widespread. More than a dozen prominent law firms have issued client advisories explicitly warning against using public AI for any matter related to legal affairs. The consensus guidance is unambiguous: users must treat public AI platforms as non-confidential environments and assume anything shared could be disclosed.
Firms like Orrick and Crowell & Moring recommend using only private, closed AI systems with strict confidentiality guarantees in their terms of service. They also stress that any use of AI for legal work should occur only under the explicit direction of an attorney. Some firms, such as Sher Tremonte, have taken the extraordinary step of updating their client engagement agreements. Their new language states that sharing a lawyer’s advice or communications with a chatbot could result in a complete waiver of attorney-client privilege.
This landmark ruling establishes a critical boundary in the digital age. While AI tools offer powerful capabilities, they do not function as legal counsel. For anyone navigating legal challenges, the safest path remains direct consultation with a qualified attorney, ensuring all strategic discussions remain within a truly privileged relationship.
(Source: The Next Web)
