About

New Evidence Shows Chatbot Conversations Are Logged, Stored, and Used to Train AI Models

Published
Score
20

Why it matters

AI chatbots retain and use user conversations by default, with no legal privilege protecting those exchanges from discovery or law enforcement access. A Stanford HAI study confirmed that all six major AI providers—including OpenAI, Meta, and Google—log user inputs for model training by default, with some retaining data indefinitely. The issue moved from theoretical to concrete when a federal arson trial involving the 2025 Palisades fire introduced a defendant's ChatGPT logs as evidence, establishing that AI conversations can be subpoenaed and admitted in court like any other digital record.

OpenAI currently operates under a court order requiring indefinite retention of chat logs to preserve evidence in ongoing cases. Meta disclosed in a class-action lawsuit that it retained raw user logs for up to 90 days, including sensitive data such as Social Security numbers and mental health crisis statements. Legal experts warn that AI chats lack the confidentiality protections afforded to attorney-client or doctor-patient communications, yet many users assume their conversations are private or ephemeral.

Attorneys should counsel clients that inputs to AI tools are archived as identifiable records tied to device IDs, IP addresses, or account profiles—and are discoverable in civil and criminal litigation. Recent research also shows that LLMs can infer sensitive personal information like race, location, and occupation with 85–95% accuracy from routine conversations. The Palisades trial precedent signals that courts will treat AI chat logs as standard digital evidence. Professionals handling sensitive matters should assume any information shared with a chatbot may eventually be produced in discovery or used against their interests.

Sources

mail Subscribe to Privacy email updates

Primary sources. No fluff. Straight to your inbox.

Also on LawSnap