ChatGPT logs surface as evidence in US court cases

Conversations with ChatGPT are increasingly appearing in US court proceedings, exposing personal disclosures, litigation discussions and potentially incriminating searches to judges, lawyers and, in some cases, the public.

A Washington Post investigation published Aug. 27 identified at least a dozen civil and criminal cases from the previous two years in which courts referenced chatbot logs, most of them involving ChatGPT. The records reached legal proceedings through subpoenas, evidence collected from devices and filings submitted by litigants.

The cases highlight a gap between how private an AI conversation may feel and how it is treated under the law. A general-purpose chatbot is not a lawyer, doctor or licensed therapist, and communications with it do not automatically receive the confidentiality protections that can cover certain professional relationships. Existing chat records also generally do not acquire Fifth Amendment protection merely because their contents could be incriminating; the constitutional privilege is principally concerned with compelled testimonial self-incrimination.

In one civil case, defense lawyers filed October 2024 ChatGPT conversations belonging to a teenager identified as R.K.C. The teenager was involved in litigation alleging that major social media companies caused addiction-related harm. The submitted material included discussion of personal matters and a possible $1 million settlement, demonstrating how chatbot use connected to an active dispute can become part of discovery or motion practice.

Criminal investigations present a separate route for conversations to enter evidence. In Missouri, investigators recovered from a suspect’s device a ChatGPT query concerning whether responsibility for vandalism could be detected. Other cases have involved users discussing violent conduct or asking chatbots about concealing crimes. Depending on the circumstances, such records can be collected during a device search, obtained from a service provider through legal process or voluntarily supplied by a party.

Whether a log is admitted at trial remains a distinct question from whether it can be obtained during discovery or an investigation. Courts can still consider relevance, authenticity, hearsay rules and the legality of the method used to acquire the record. The treatment of AI-assisted legal work is also unsettled. Lawyers may argue that some material generated while preparing litigation is protected as attorney work product, but that protection depends on the facts and does not create a broad privilege for ordinary chatbot conversations.

The expanding use of AI records in litigation also intersects with disputes over how long providers retain data. In May 2025, a federal magistrate directed OpenAI to preserve user conversations, including chats users had deleted, in connection with copyright litigation. That preservation process later resulted in the production of 20 million de-identified conversation logs in early 2026. De-identification can reduce direct exposure of personal information, but it does not resolve the broader question of when stored conversations may be subject to court orders.

Government demand for account information provides another measure of legal interest in chatbot data. OpenAI reported receiving 75 government requests for user content during the second half of 2025. The company disclosed content in response to 62 of those requests, covering 84 accounts.

AI chat logs fit within the wider category of electronically stored information, alongside emails, direct messages, cloud documents and search histories. Their conversational design, however, can encourage unusually candid disclosures. Users may describe fears, legal strategies, health concerns or alleged misconduct in natural language without recognizing that the resulting transcript could be retained on a device or server.

Courts have not established a single nationwide rule for every use of chatbot evidence, particularly when lawyers employ AI tools while preparing a case. For ordinary users, the emerging record is clearer: typing information into a chatbot does not, by itself, make the exchange legally confidential or prevent it from later being sought in litigation.

Sources: AI litigation