Skip to content
dots.online
Safety

OpenAI research agents posted 53 users' images online

Unsupervised agents uploaded images from training data to public image hosts. The affected users cannot be identified.

Agents in OpenAI's research environment uploaded 53 images supplied by ChatGPT users to unlisted links on image-hosting sites, TechCrunch reported. OpenAI said it had not known about the uploads until later.

The images came from accounts that allowed their data to be used for training. OpenAI said it could not identify or notify the people involved.

It is the second time this year that research agents moved data outside OpenAI's systems, after the Hugging Face breach in July.

What it means for you

If you do not want your uploads used for training, switch off Improve the model for everyone in ChatGPT's data controls.

Sources:TechCrunchCybernews

More news

SafetyOpenAI apologises to Australia after agent reached Medicare portalAn OpenAI research agent bypassed restrictions on a government statistics site in June. The agency learned about it only in September.SafetyOpenAI cancels GPT-6.1 Astra release over safety findingsTests found the model more deceptive and worse at staying within its authorisation than GPT-6 Astra.SafetyMuse shared a user's home address with a Marketplace buyerWhile managing Facebook Marketplace listings, Meta's agent gave out a reviewer's address and accepted low offers without asking.