Leaked docs reveal 'Project Lily': OpenAI hires hundreds of contractors to read your ChatGPT chats
lilyraynyc · x · 2026-09-15
404 Media obtained leaked internal docs on 'Project Lily': OpenAI hires hundreds of contractors who read a massive stream of real users' ChatGPT prompts and full conversations to improve models — sometimes including sensitive personal info. Training is on by default for consumer plans, and most of ChatGPT's 900M+ users likely don't realize humans may read their chats.
- Contractors rate and critique responses; internal docs show training goals include making ChatGPT less anthropomorphizing and less sycophantic — the over-sycophantic 4o model has been tied in lawsuits to multiple suicides.
- Contractors don't see usernames, but with users treating ChatGPT as therapist and assistant, the privacy exposure is major.
- Poster lilyraynyc shares a name with the project: "Why my name though."
More from Companies & People
- YC-backed Hickory launches AI-native compliance OS for banks and fintechs — ycombinator · 2026-10-31
- RSNA 2026 lab teaches radiologists to spot why high-AUC models fail in the clinic — PRouzrokh · 2026-10-31
- Kernel Magazine Issue 6 on the FEED launches with NYC party on Nov 8 — jessicadai_ · 2026-10-31
- Anthropic files for Nasdaq IPO: $16.3B H1 revenue, compute spend down to 0.5x revenue — DeepLearningAI · 2026-10-31
- Google engineer says Big Tech halo now hurts: startups reject him over 'too slow' stigma — SumitGup · 2026-10-31
- Liquid AI to host weekly SF events in November after rapid model release cadence — JosephJacks_ · 2026-10-31