Breaking

OpenAI Hired Hundreds of Contractors to Review ChatGPT Chats Under Project Lily

September 14, 2026 at 21:29 EDT

Policy & Regulation
Security
Foundation Models

OpenAI has hired hundreds of outside contractors to review real ChatGPT conversations under an internal effort called Project Lily, intensifying scrutiny of how consumer AI chats are used to improve foundation models. The reviewers do not see ChatGPT usernames, but the work can involve full conversations and sensitive context after automated filtering.

September 14, 2026 · OpenAI

Human Review Enters the Privacy Spotlight

Project Lily uses contractors to review real ChatGPT conversations, with usernames stripped and a privacy filter applied before review. The issue is where model improvement ends and user expectations of privacy begin.

Hundreds

contractors reviewing chats

900M

weekly active ChatGPT users

96%

Privacy Filter F1 score

128k

token context window

Retention Contrast

30 days

ChatGPT Temporary Chat

Up to 3 years

reviewed Gemini chats

Each blue block represents one month; the orange block represents one 30-day period.

How the Review Loop Works

User conversation

→

Privacy Filter

→

Contractor rating

→

Model improvement

Why teams use it

Human reviewers can catch tone, intent, and safety failures that automated metrics may miss.

Why users worry

Filtered chats can still carry sensitive context, and privacy controls are easy to miss.

The unresolved question

The technology reduces exposure but does not eliminate it. The policy test is whether users understand review, retention, and opt-out choices before sharing sensitive information.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.