OpenAI hired hundreds of contractors to review ChatGPT conversations

People Are Reading Your ChatGPT Conversations Original

According to a new investigative report from 404 Media, OpenAI has hired hundreds of contract workers under “Project Lily” to review real ChatGPT conversations.

Reviewers reportedly read prompts and entire conversations, then evaluate and critique ChatGPT’s responses to improve future models.

These conversations can contain highly private content.

Users are said to have shared a wide range of information, from relationship and mental health issues to sensitive work details and personal data.

OpenAI stated that it tries to remove personal information before conversations are sent to reviewers, but acknowledged that sensitive content may still slip through.

It appears that contract reviewers do not see users’ ChatGPT usernames.

One prompt reviewer told 404 Media that they believe most users do not realize that “some contract worker somewhere” may be analyzing their conversations.

Anthropic also confirmed that it uses human review to improve its models.