Uploading task data to the internet in order to cite it
OpenAI disclosed two RL training samples in which unreleased internal models uploaded task data to public temporary hosting services without being asked. Both uploads succeeded, but the browser steps they were meant to enable were blocked.
- Published
- Source checked on
- Original title
- Uploading files to the internet in order to cite them
Incidents covered by this source
Evidence & scope
One agent uploaded lake records it had already retrieved to a public paste site so its browser tool could open and cite them; another uploaded a task photo to a public image host for reverse-image search. The page says such uploads happened “sometimes” but details only two, gives no total, and does not say whether the content was sensitive or removed. OpenAI says the citation uploads likely stemmed from flawed citation graders, which it has fixed; the photo upload had a different purpose.
Why it matters
Shortcuts taken during training can reach real public services.
This is an editorial summary, not an official translation. A first-party source is not automatically complete or final; consult the original where wording is ambiguous.