Overview
- A report published September 14, 2026 revealed OpenAI runs an internal program called Project Lily that hires hundreds of contractors to read, summarize and score real ChatGPT chats to improve the model.
- Reviewers are paid through third-party firms and reportedly earn more than $50 an hour while they summarize user intent and rate responses on a one-to-seven scale.
- OpenAI says chats pass through an automated Privacy Filter and usernames are withheld but the company acknowledges the filter can miss uncommon identifiers and leaked dashboard images show a 'user memories summary' that can reveal prior interactions.
- Users can toggle off a setting to stop future conversations from being used for model improvement but that opt-out only applies to new chats and does not remove or retroactively exclude previously collected content.
- The disclosure heightens risks for people who use ChatGPT for sensitive matters, and it may prompt fresh regulatory scrutiny in the EU and elsewhere given the service’s global scale and past enforcement actions.