OpenAI is hiring hundreds of contractors to read live ChatGPT chats
404 Media reports OpenAI is hiring hundreds of contractors under Project Lily who read real ChatGPT prompts and score replies. Usernames are stripped; OpenAI said sensitive details can still get through. The Next Web adds that the company would not say where it tells users humans may read chats, and that Europe's highest court last September put the duty to inform on the collector.
Artificial Intelligence··Night
Hundreds of reviewers, usernames off
404 Media, which saw internal documents, reports that OpenAI is hiring hundreds of contractors who read a large stream of real ChatGPT prompts, sometimes whole conversations, to improve replies. Contractors train the model not to anthropomorphize itself and to be less sycophantic. They do not see ChatGPT usernames. OpenAI said it tries to remove personal information before prompts reach reviewers and acknowledged that sensitive details can still get through. Anthropic told 404 Media it uses human reviewers on Claude for users who have Help improve our AI models turned on.[1]
Where users are told remains unanswered
The Next Web, citing the same 404 Media investigation, asked OpenAI where it tells users that humans may read their chats. The company did not answer before publication and later pointed to a help page. Reviewers can see a memories summary above the prompt, which sometimes shows prior use and a possible location. OpenAI's Privacy Filter page says the model can miss uncommon identifiers. Improve-the-model is on by default for free, Plus and Pro accounts and off for Enterprise, Business and Edu; switching it off applies to new conversations, not old ones.[2]
The collector's duty, not the contractor's
The Next Web notes the Court of Justice held last September in EDPS v SRB that a controller's duty to inform applies before any transfer and is judged from the controller's point of view. Whether a North American contractor could identify the writer is not the test. Italy's regulator had already fined OpenAI 15 million euros, 9 million of it for processing without an adequate legal basis, and ordered public-information advertising. 404 Media's documents describe the review programme; they do not themselves recast European case law.[2], [1]
