OpenAI Contractors Read User ChatGPT Chats Under Project Lily
The setup leaves a gap between what the Privacy Filter claims to strip and what reviewers can still see, and OpenAI's own explanation concedes the filter may miss context-dependent identifying details.
Reporting from 1 source: GIGAZINE.
OpenAI runs an internal project called Project Lily in which several hundred contract staff read real conversations between users and ChatGPT and evaluate response quality. Usernames are hidden and a Privacy Filter replaces names, addresses, email addresses, and phone numbers with labels, but the filter cannot remove everything. OpenAI says it may miss uncommon identifying information or details whose meaning depends on context. ChatGPT for individuals uses conversations for model improvement by default.
OpenAI's Privacy Filter swaps names, addresses, email addresses, and phone numbers for labels before a conversation reaches a reviewer, but the company states it may overlook uncommon identifying information or details whose meaning changes with context. That second category is the hard one, since a sentence can point at a person without naming them. The review screen also anonymizes usernames, yet the memory summary ChatGPT keeps can appear, and according to The Next Web that summary may carry past usage and approximate residential area. ChatGPT for individuals is set by default to use conversations for model improvement; Business, Enterprise, and Edu conversations are not, and Temporary Chat is not either. Turning off "Help improve the model" under Data controls stops new conversations from being used for training.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.