conv.

All stories
AIQuiet 6d · day 10

404 Media reveals OpenAI contractors read sensitive ChatGPT conversations

Leaked documents show hundreds of human reviewers analyze real user chats including personal details to improve the model.

newest coverage 6d agosummary written 9d ago51 pieces816 likes & upvotes11 comments MastodonNewswiresHacker NewsRedditXBlueskyGoogle News

Part of a larger narrative

OpenAI Security and Privacy Crisis

2 stories · since Sep 14 · newest 2h ago — OpenAI faces mounting scrutiny over data handling practices as revelations about contractor access to sensitive conversations collide with the company's…

  1. ChatGPT Voice gets plugin access and GPT-6 models
  2. 404 Media reveals OpenAI contractors read sensitive ChatGPT conversationsyou are here

What to know

  • OpenAI employs hundreds of human contractors who read real ChatGPT user prompts and conversations to improve model responses, a practice unknown to most of the platform's 900+ million users.
  • Sensitive personal information reaches reviewers despite company claims to anonymize and filter data before human review; some user prompts explicitly request privacy, suggesting users do not expect human access.
  • The practice, dubbed "Project Lily" internally, aims to reduce ChatGPT's sycophancy and anthropomorphization; Anthropic confirmed it operates a similar human review system.
most voices

This is a significant privacy concern that users should know about.

  • “OpenAI has hired an army of contractors who read real ChatGPT users' chats. I've seen internal docs, the review system, and real user prompts. Sometimes they contain very personal and sensitive information”

    josephcox@infosec.exchange · Mastodon ↗
some voices

Human review is a necessary and often overlooked part of AI training and model improvement.

  • “Let's not forget that, in part, humans train AI”

    arjen@idf.social · Mastodon ↗

“OpenAI is hiring hundreds of contractors who read a massive stream of real users' ChatGPT prompts, with the prompts sometimes including sensitive personal information.”

404 Media, Publication · 404 Media ↗

OpenAI AI company operating ChatGPT404 Media Investigative publicationAnthropic AI company

404 Media reveals OpenAI contractors read sensitive ChatGPT conversations
404media.co

How it unfolded 1 development · click the chart to see its coverage articlesposts

Peak 22 pieces in 3h at Sep 14, 9 AM; 51 pieces over 10 days (16 articles · 35 posts) Sep 14, 9 AM — 22 pieces · 7 articles · 15 posts — Mastodon 9, Newswires 6, Bluesky 3, +3 moreSep 14, 12 PM — 7 pieces · 1 article · 6 posts — Mastodon 6, Newswires 1Sep 14, 3 PM — 1 piece · 1 post — Mastodon 1Sep 14, 6 PM — quietSep 14, 9 PM — 1 piece · 1 post — Mastodon 1Sep 15, 12 AM — 1 piece · 1 post — Mastodon 1Sep 15, 3 AM — 1 piece · 1 post — Mastodon 1Sep 15, 6 AM — 2 pieces · 1 article · 1 post — Newswires 1, Mastodon 1Sep 15, 9 AM — 1 piece · 1 post — Mastodon 1Sep 15, 12 PM — 5 pieces · 5 articles — Google News 5Sep 15, 3 PM — 1 piece · 1 post — Mastodon 1Sep 15, 6 PM — quietSep 15, 9 PM — 1 piece · 1 post — Mastodon 1Sep 16, 12 AM — 2 pieces · 2 posts — Reddit 1, Mastodon 1Sep 16, 3 AM — quietSep 16, 6 AM — quietSep 16, 9 AM — 3 pieces · 2 articles · 1 post — Mastodon 2, Newswires 1Sep 16, 12 PM — quietSep 16, 3 PM — quietSep 16, 6 PM — quietSep 16, 9 PM — 1 piece · 1 post — Mastodon 1Sep 17, 12 AM — quietSep 17, 3 AM — quietSep 17, 6 AM — quietSep 17, 9 AM — quietSep 17, 12 PM — quietSep 17, 3 PM — 2 pieces · 2 posts — Mastodon 2Sep 17, 6 PM — quietSep 17, 9 PM — quietSep 18, 12 AM — quietSep 18, 3 AM — quietSep 18, 6 AM — quietSep 18, 9 AM — quietSep 18, 12 PM — quietSep 18, 3 PM — quietSep 18, 6 PM — quietSep 18, 9 PM — quietSep 19, 12 AM — quietSep 19, 3 AM — quietSep 19, 6 AM — quietSep 19, 9 AM — quietSep 19, 12 PM — quietSep 19, 3 PM — quietSep 19, 6 PM — quietSep 19, 9 PM — quietSep 20, 12 AM — quietSep 20, 3 AM — quietSep 20, 6 AM — quietSep 20, 9 AM — quietSep 20, 12 PM — quietSep 20, 3 PM — quietSep 20, 6 PM — quietSep 20, 9 PM — quietSep 21, 12 AM — quietSep 21, 3 AM — quietSep 21, 6 AM — quietSep 21, 9 AM — quietSep 21, 12 PM — quietSep 21, 3 PM — quietSep 21, 6 PM — quietSep 21, 9 PM — quietSep 22, 12 AM — quietSep 22, 3 AM — quietSep 22, 6 AM — quietSep 22, 9 AM — quietSep 22, 12 PM — quietSep 22, 3 PM — quietSep 22, 6 PM — quietSep 22, 9 PM — quietYesterday, 12 AM — quietYesterday, 3 AM — quietYesterday, 6 AM — quietYesterday, 9 AM — quietYesterday, 12 PM — quietYesterday, 3 PM — quietYesterday, 6 PM — quietYesterday, 9 PM — quietToday, 12 AM — quiet 1
Sep 15Sep 16Sep 17Sep 18Sep 19Sep 20Sep 21Sep 22now · 3:45 AM ET
  1. 1

    Anthropic confirmed to use human review for model improvement

    404 Media reported that Anthropic confirmed it also employs human reviewers to improve its models, indicating the practice is industry-wide rather than specific to OpenAI.

    “No, I don't think they would imagine some contractor somewhere [...] is analyzing the conversations.”
    — OpenAI human reviewer, Prompt reviewer · source

    • This article makes two important points: 1. People using ChatGPT expose sensitive data because they think it is private 2. OpenAI relies on hundreds of human contractors reviewing replies to improve their models https://www. 404media.co/inside-project-lil y-the-humans-reading-your-chatgpt-chats/

      evacide@hachyderm.ioMastodon9d ago371▲view on Mastodon ↗
    2 more of the top 3 · 20 posts in this stretch

    • There is a lot of new info in this article, based on documents, Slack chats, instruction guides, and me seeing real ChatGPT user prompts in the review system. Firstly, a worker we spoke to thinks ChatGPT users have no idea their chats may be read by someone else www.404media.co/inside-proje... …

      @josephcoxBluesky9d agoview on Bluesky ↗
    • mercedesallen@mastodon.social

      "You don't have to go very far beneath the surface — beneath the mirror — to find that not only are these things built in the image, but usually a fairly bad facsimile thereof, of what human abilities can do. But they require constant, constant intervention from humans…" Inside ‘Project Lily’: The Humans Reading Your ChatGPT Chats https://www…

      mercedesallen@mastodon.socialMastodon9d ago1▲view on Mastodon ↗
    all of them →
  2. background

    OpenAI acknowledges sensitive personal data reaches human reviewers despite safeguards — The company attempts to anonymize prompts (removing usernames) and claims it tries to strip personal information before contractors review them, but acknowledged that sensitive details still get through. Some user prompts explicitly ask ChatGPT to keep information private, suggesting users do not expect human review.

  3. background

    Contractors rate ChatGPT responses across three-stage review process — Human reviewers access a dashboard where they select tasks and view real user prompts without usernames. They then summarize the user's intent and rate ChatGPT-generated responses, training the model to reduce anthropomorphization and sycophancy.

  4. background

    404 Media discloses OpenAI's "Project Lily" human review operation — 404 Media published an investigation based on leaked internal documents, instruction guides, Slack channels, and real user prompts, revealing that OpenAI has hired hundreds of contractors to read and review ChatGPT conversations to improve model responses.

Also covered reported alongside — the timeline has no entry for these yet

  1. 4 outlets Humans may be reading your ChatGPT prompts — here's how to stop it

    first by Tom's Guide, 9d ago · also The Decoder, India Today, 404 Media

    3 more headlines
  2. 2 outlets Hundreds of contractors can read ChatGPT chats containing personal data: Report | 'The practice is part of 'Project Lily'' | Inshorts

    first by Inshorts, 8d ago · also 404 Media

    2 more headlines
  3. 1 outlet ChatGPT transcripts are reportedly read by humans to improve responses, including those with personal information …

    first by Tom's Hardware, 9d ago

    1 more headline

and 1 smaller piece

What people are saying 12 voices from 2 sites · best of 20 · verbatim