Project Lily Exposed: OpenAI Contractors Manually Read and Score Real User Chats, Says Report
OpenAI hired hundreds of contractors for 'Project Lily' to manually read and score real ChatGPT chats to improve tone. While filters remove names, sensitive details remain visible. Users can opt out via data settings. That said, it does not guarantee complete immunity from human review for safety, security, or legal requirements.
OpenAI has engaged hundreds of external contractors to manually read and evaluate real-world ChatGPT conversations as part of an internal quality-control initiative, bringing renewed scrutiny to user data privacy practices. The program underscores the heavy reliance on human labour behind the scenes to fine-tune automated artificial intelligence models.
What Is 'Project Lily'?
According to internal documents obtained by 404 Media, "Project Lily" is an initiative where contractors are hired to manually review, score, and grade actual conversations between users and ChatGPT. The leaked materials reveal that reviewers examine full conversation threads, summarise user goals, compare model-generated answers, and grade responses based on usefulness, tone, and writing quality. ChatGPT Hallucination Case: US Lawyer Fined USD 5,000 Over Fake Witness Testimony.
The evaluation process is specifically designed to root out behavioural flaws, such as excessive agreement with users, patronising phrasing, unnatural "AI-speak", unnecessary emojis, and responses that mistakenly imply the chatbot possesses human emotions or personal experiences.
Privacy Considerations and Sensitive Data Handling
While OpenAI utilises automated privacy filters to strip usernames and identifying details before chats reach human reviewers, the filtering process is not flawless. Reviewers may occasionally view sensitive personal disclosures concerning work, finances, or relationships that users might assume remain strictly between themselves and the AI. This dynamic has prompted questions from contractors regarding whether ordinary users fully realise that real people can access and evaluate their chat logs. What Is the Viral 80s AI Photo Trend? Step-by-Step Guide to Creating Your Own Retro Images Using ChatGPT.
Controlling Your Data and Opt-Out Settings
Users who prefer to restrict how their conversations are handled can disable data sharing on personal accounts by navigating to Settings > Data Controls > Improve the model for everyone. While turning off this setting stops future chats from being used for model training, it does not guarantee complete immunity from human review for safety, security, or legal requirements. Alternatively, users can select the "Temporary Chat" feature to prevent sessions from being stored in history or utilised for model improvement.
DISCLAIMER: This article is strictly for informational and educational purposes to raise awareness about the said topic. LatestLY DO NOT host, distribute, or provide links to sexually explicit content, leaked videos, or pirated material. Any search terms or keywords mentioned are solely for the purpose of explaining the viral trend and its associated dangers. We strongly condemn the circulation of non-consensual intimate imagery (NCII) and advise all readers to refrain from searching for or sharing such content, as doing so may lead to severe legal consequences.
(The above story first appeared on LatestLY on Sep 15, 2026 10:14 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).
