That late-night ChatGPT session where you asked something you’d never Google? It might end up in a courtroom.
A Washington Post investigation published on August 27 found that at least a dozen US civil and criminal court cases over the past two years have referenced chatbot conversation logs, primarily from ChatGPT. The logs have been subpoenaed, submitted as evidence, and entered into the public record, exposing deeply personal content that users almost certainly assumed would stay between them and their favorite AI.
Your AI therapist will testify against you
The core issue is deceptively simple. ChatGPT conversations are treated by courts as standard electronic records. They carry no special protection. No attorney-client privilege. No therapist-patient confidentiality. No Fifth Amendment shield.
One of the more prominent cases involves R.K.C., a teenager who was part of litigation against major social media companies over alleged addiction-related harms. Defense attorneys submitted R.K.C.’s October 2024 ChatGPT conversations in court filings. Those logs included sensitive discussions about a potential $1 million settlement and personal issues, the kind of material that would normally be discussed behind closed doors with a lawyer, not typed into a chatbot.
The criminal side is even more stark. In a Missouri case, law enforcement retrieved a suspect’s ChatGPT query about detecting responsibility after committing vandalism. The query was pulled directly from his device. In other instances, chat logs revealed individuals asking AI how to cover up crimes or discussing violent acts, which triggered law enforcement alerts.
OpenAI’s data vault
The legal machinery around AI chat logs has been building quietly. In May 2025, a federal magistrate ordered OpenAI to preserve user logs, including deleted chats. That order led to a significant production: 20 million de-identified conversation logs were handed over in early 2026 as part of ongoing copyright litigation.
OpenAI’s own transparency report for the second half of 2025 offers a window into government interest in user data. The company received 75 government content requests during that period and complied in 62 cases involving 84 accounts.
The legal gray zone
Court rulings on AI chat log admissibility have been inconsistent. Some courts have treated AI conversations as straightforwardly discoverable electronic evidence, no different from emails or text messages. Others have grappled with whether certain AI-assisted work might qualify for work-product protections, particularly when attorneys use chatbots during litigation preparation. The results have been mixed, creating a patchwork of precedents that varies by jurisdiction.