OpenAI, the American company behind the widely used ChatGPT platform, has introduced a new initiative called "Project Lily," according to 404 Media, an American news website. The project involves paying hundreds of third-party providers to review conversations between users and ChatGPT in order to improve the AI model. However, users are reportedly not informed about this process, as OpenAI claims to have billions of users worldwide. To protect user privacy, OpenAI says that all conversations pass through a system called Privacy Filter, which is an AI tool designed to automatically "mask personally identifiable information" (PII), such as names, addresses, or other sensitive details. This process aims to ensure that personal data is not shared with the third-party reviewers. However, OpenAI acknowledges that the Privacy Filter is not perfect and can sometimes fail to identify uncommon names, unique identifiers, or ambiguous private references. Additionally, the Privacy Filter may not always accurately determine how much information to mask, especially when the context of a conversation is limited or unclear. This could lead to situations where some personal data might still be exposed to the third-party reviewers, even if unintentionally. Users who wish to prevent their future conversations from being used for model improvements can disable the "Improve the model for all" option. This setting is found in the "Data Management" submenu, which is accessible by clicking on the user's name in the bottom left corner of the ChatGPT interface. It’s important to note that disabling this option does not erase past conversations but stops future interactions from being used in the training process.