OpenAI Contractors Review ChatGPT Conversations Despite Opt-Out Settings

Malwarebytes reported on 15 September 2026 that 404 Media had found OpenAI hiring hundreds of contractors to review real users’ ChatGPT prompts and responses under a programme reportedly called “Project Lily”. Contractors score or critique answers to help improve the chatbot. OpenAI says a limited number of authorised staff and trusted service providers may access chats for abuse investigations, support, legal matters and model improvement.

It told 404 Media that a “Privacy Filter” removes personal information before review, but acknowledged that sensitive details can still pass through. Although reviewers do not see users’ names, they may see personal information in conversations and a “user memories summary”, which could reveal substantial information about an individual.

Users can opt out of ordinary model training, but this does not prevent all human access. In ChatGPT, the “Improve the model for everyone” setting is enabled by default for Free, Plus and Pro plans; users can disable it under Settings > Data Controls. Perplexity users can turn off AI data retention under Settings > Preferences, although previously collected data is not removed and information may still be processed to operate or improve the service or meet legal obligations.

Claude users can disable “Help Improve our AI Models” under Settings > Privacy. Anthropic says this excludes new and previously stored conversations from future training, but data already included in training cannot be removed. Conversations flagged by safety classifiers may still be used for trust-and-safety models, policy enforcement or safety research.