OpenAI reads users' ChatGPT conversations

According to a 404 Media investigation, OpenAI hired hundreds of contractors as part of Project Lily to read real ChatGPT user conversations and rate model responses on a scale of 1 to 7, aiming to reduce sycophancy and excessive anthropomorphism.

Compiled by Sergey KostenchukPublished 2026-09-16Updated 2026-09-18

2026-09-16 Safety OpenAI

🛡 OpenAI reads users' ChatGPT conversations

According to a 404 Media investigation (September 14, 2026, leaked internal documents), OpenAI is hiring hundreds of contractors for Project Lily: they read real ChatGPT user chats and rate model responses on a scale of 1 to 7, training the bot to avoid sycophancy and anthropomorphism.

🌍 Crossing Hurdles hires the reviewers, and Mercor pays them — at a rate above $50 an hour. Dialogues pass through a privacy filter, but OpenAI acknowledges: sensitive details slip through, and even requests to keep things confidential were visible to reviewers.

👤 If you haven't disabled “Improve the model for everyone” (Settings → Data Controls; also applies in Codex), your chats may be read by humans. Temporary chats are not used for training and are deleted after 30 days.

Source 1: https://www.404media.co/inside-project-lily-the-humans-reading-your-chatgpt-chats/ Source 2: https://help.openai.com/en/articles/7730893-data-controls-faq

Sources