{"id":409,"date":"2026-09-16T08:07:45","date_gmt":"2026-09-16T00:07:45","guid":{"rendered":"https:\/\/aidashxp.com\/openai-project-lily\/"},"modified":"2026-09-16T08:07:45","modified_gmt":"2026-09-16T00:07:45","slug":"openai-project-lily","status":"publish","type":"post","link":"https:\/\/aidashxp.com\/en\/openai-project-lily\/","title":{"rendered":"OpenAI Reveals \u201cProject Lily\u201d: Outsourced Contractors Read Real Conversations from 900 Million Users"},"content":{"rendered":"

If you\u2019ve used ChatGPT<\/a> sensitive information, this news warrants your immediate attention. According to 404 Media<\/strong> reports, OpenAI internally runs a project codenamed Project Lily<\/strong> \u2014hundreds of outsourced workers directly read real users\u2019 ChatGPT<\/a> prompts (including sensitive personal information) and score the model\u2019s responses to reduce tendencies toward \u201csycophancy\u201d and \u201cover-personification.\u201d<\/p>\n\n\n\n

Who\u2019s Reading Your Conversations<\/h2>\n\n\n\n

Reports indicate that these outsourced workers were recruited by Crossing Hurdles and compensated via the Mercor platform, with some roles paying over $50 per hour. ChatGPT<\/a> Currently has Over 900 million users<\/strong>The trigger for all this is a setting named \u201cImprove the model for everyone\u201d<\/strong> whichEnabled by default<\/strong>must be manually disabled.<\/p>\n\n\n\n

The issue is not review itself, but scale and default behavior.<\/h2>\n\n\n\n

To be fair, human review of anonymized transcripts has long been standard practice across AI labs. What is truly alarming are three things: scale (900 million users), deliberate obfuscation via codenames, and the \u201copt-in by default\u201d design\u2014which means most users have no idea this toggle even exists. Adding irony, OpenAI publicly endorsed the \u201ctransparency reporting\u201d provision in the FRONTIER Act just that same week\u2014yet such reports should have disclosed Project Lily before journalists uncovered it.<\/p>\n\n\n\n

Anonymization does not equal safety<\/h2>\n\n\n\n

Another unsettling detail from the report is that, even after \u201canonymization,\u201d these transcripts routinely contain names, addresses, medical records, and other personally identifiable information in practice. For ordinary users, this means anything you paste into ChatGPT<\/a> could become a data point on an outsourced worker\u2019s task list. Especially concerning is that many people use ChatGPT<\/a> as a \u201cpersonal assistant\u201d without realizing the free tier participates in this data return by default.<\/p>\n\n\n\n

What you should do<\/h2>\n\n\n\n