Investigation Reveals Hundreds of Contractors Reading Real ChatGPT Conversations
View original source →A new investigation revealed that OpenAI has hired hundreds of contractors, working under an internal effort called "Project Lily," to read real users' ChatGPT conversations—including sensitive material—in order to improve how the model responds.
Key points:
• Contractors review live user prompts and model replies to flag problems and refine future model behavior; OpenAI says contractors don't see usernames and that it strips personally identifying information beforehand
• OpenAI acknowledged that despite its filtering efforts, "sensitive details can still get through" to the humans doing the review
• The investigation raises the practical question of what "private" means when using a consumer AI chatbot for anything containing personal, medical, financial, or legal detail
For any regulated organization—including insurance and financial services—this is a concrete reminder that consumer-grade AI chatbots are not confidential channels by default.
Why It Matters: Consumer ChatGPT accounts are not confidential channels. Compliance teams should confirm which AI tools in use are enterprise-tier with contractual no-training guarantees versus consumer-tier like this.