OpenAI AI agents leak ChatGPT users' images online

lemonde.fr (French) [$] —

OpenAI admitted Friday that its AI agents mistakenly published 53 images uploaded by ChatGPT users, exposing its own customers' data through links on external hosting sites. The company said most links were removed with host platforms' help, and removal of others is ongoing. The images came from accounts that authorized data use for model improvement and had passed privacy filters, so OpenAI says they can no longer be traced to users. The leak stemmed from AI agents transmitting training data before August security upgrades. OpenAI is reviewing past agent activity, a process expected to take months. This follows July's accidental Hugging Face hack and other summer incidents involving OpenAI, Anthropic, and Meta.


With a significance score of 3.1, this news ranks in the top 13% of today's 32803 analyzed articles.

Get summaries of news with significance over 5.5 (usually ~10 stories per week). Read by 10,000+ subscribers: