404 Media reports that OpenAI's internal project codenamed 'Project Lily' has employed hundreds of outsourced prompt moderators to manually review anonymized conversation records between real users and ChatGPT. Moderators are required to assess the quality of responses, check for relevance, and screen for issues such as 'AI-like phrasing', preachy tones, and emoji usage, while ensuring that responses do not contain anthropomorphic expressions or fabricated personal experiences. The project offers an hourly wage exceeding $50, but the operational guidelines frequently change and contain contradictions. Although chat records are anonymized, the filtering process may still leak personal data, with shorter conversations posing higher risks. Additionally, user memory summaries may include user queries, interests, preferences, and even location information. OpenAI admits that it did not explicitly inform users that their records would be manually reviewed, only vaguely mentioning it in the privacy policy, sparking controversy over privacy protection.