Five types of ChatGPT queries that can get users into serious trouble

Over the past eighteen months, ChatGPT conversations have been submitted as legal evidence, indexed by Google, and retained by court order long after users deleted them. This overview looks at five categories of prompts that most often put users at risk, along with real-world cases and the regulatory response that has started to emerge.

■ Legal and confidentiality risks
- Court cases worldwide involving fabricated citations, with self-represented litigants and lawyers most often at fault
- How ChatGPT prompts can be used as evidence against you in a dispute
- OpenAI's updated usage policies on licensed professional advice

■ Workplace and account data
- Uploading confidential work documents to the public version, including a U.S. cyber agency case and Samsung's 2023 restrictions
- How consumer conversations can be used for model training, and what opting out still leaves exposed
- Risks of sharing passwords, given how the tool remembers prior conversations

■ Medical and mental health use
- A case in which dietary advice led to a hospital stay, and safety guardrails that do not cover every condition
- Chatbots reinforcing existing beliefs, and U.S. states legislating against AI-provided therapy

■ Shared conversations
- Thousands of public chats appearing in search results, and the feature changes that followed

Whether you use the tool for work, health questions, or everyday problem-solving, you will come away with a clearer sense of what belongs in a chat and what should stay with a doctor, lawyer, or therapist.