OpenAI, the company behind the popular chatbot ChatGPT, is paying hundreds of contractors to read and rate real conversations under the internal codename Project Lily. This practice has sparked concerns about user privacy, as the contractors can see a user's memories summary above the prompt, which sometimes reveals what the person has used the chatbot for before and where in the world they may live.
According to 404 Media, the reviewers do not see usernames, but can still access sensitive information. OpenAI claims that a Privacy Filter model strips personal information first, but its own page for that model says it can make mistakes, miss uncommon identifiers, and under-redact when context is limited.
The company did not answer when asked where it had informed users that humans may read their chats. After publication, OpenAI pointed to a help page, but the lack of transparency has raised eyebrows. Anthropic, another AI company, confirmed it uses human review too, but only for users who switch the setting on, and says it removes account identifiers first.
Google's Gemini carries a line saying humans review some saved chats. The Court of Justice held last September that a controller's duty to inform people applies before any transfer, and is judged from the controller's own point of view rather than the recipient's. This means that whether a contractor in North America could work out who wrote a prompt is not the test; the obligation sat with OpenAI at the moment it collected the conversation.
Italy's regulator has already fined OpenAI EUR 15M, nine of it for processing without an adequate legal basis, and ordered six months of public information advertising on Italian television and radio. The default setting matters, as 'Improve the model for everyone' is switched on for free, Plus, and Pro accounts, and off for Enterprise, Business, and Edu accounts. Switching it off applies to new conversations rather than old ones.
OpenAI has spent the year selling privacy to the other half of its market, previewing zero data retention for enterprise customers in August. However, consumers get the setting that feeds the model, which has led to criticism. 'I don't think they would imagine some contractor somewhere is analyzing the conversations,' one reviewer said. The same company has been asking those users to connect their bank accounts, further fueling concerns about data protection.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.