Hundreds reading your chats

OpenAI employs hundreds of contractors to read user prompts, including messages with sensitive personal information, to improve ChatGPT's responses.

Not for safety, for improvement

This isn't a moderation effort. Contractors rate and critique ChatGPT's replies to nudge it away from acting human-like and overly agreeable.

Users don't know it's happening

When asked if ChatGPT users knew humans were reading their chats, one contractor said no. Most people would find this surveillance alarming if they knew.

“No.”

Privacy settings buried by default

OpenAI allows users to opt-out of human review through an "improve the model for everyone" setting, but it's enabled by default and not clearly disclosed.

Personal details still visible

Even though prompts are anonymized, the contractor dashboard includes a user memories summary that can reveal where someone lives and other personal details.

Chatbots create false intimacy

Experts note that chatbot interfaces make people feel like they're in private conversations, unlike social media where moderation is expected.

“chatbot interfaces automatically create a false sense of intimacy and privacy.”