OpenAI has hundreds of contract workers reading your ChatGPT conversations
| Source: THE DECODER
Tags: OpenAI, ChatGPT, privacy, RLHF, human feedback, data collection, consent
OpenAI employs hundreds of contract workers — recruited through Crossing Hurdles and paid via Mercor at 0+/hour — to read and rate real ChatGPT conversations on a 1–7 scale to reduce sycophancy. The opt-in setting is on by default; disclosure exists only in a buried FAQ, per a 404 Media investigation.
Details
OpenAI uses hundreds of contract workers to review real ChatGPT conversations, according to a 404 Media investigation based on leaked internal documents and worker interviews. Reviewers are recruited through Crossing Hurdles and paid via Mercor, earning over $50/hour. Their job: rate ChatGPT responses on a 1–7 scale to reduce excessive flattery and human-like behavior in the model. The prompts are anonymized before review, but OpenAI acknowledges its privacy filter 'can make mistakes.' Some reviewed conversations included users explicitly asking ChatGPT to keep the contents private — unaware any human could be reading. The consent mechanism is the 'Improve the model for everyone' setting, enabled by default for all users. Turning it off only protects future conversations. Temporary chat mode offers separate protection. OpenAI's only disclosure is a FAQ page online since 2023, which 404 Media argues is too vague to constitute informed consent. The practice is standard across labs: Anthropic confirmed it runs human review under a similar opt-in default, anonymizing chats by stripping email addresses. Google does the same with Gemini. The substantive question is whether opt-in-by-default plus a buried FAQ qualifies as meaningful transparency.