OpenAI reports image leaks and data leaks by agents

According to OpenAI, the images were published on hosting sites through links that were not publicly listed.

Most of the images have already been removed with the help of the hosting services involved, and the company says it is working to remove the rest.

The images had been uploaded by users who allowed their interactions with ChatGPT to be used to train OpenAI models. Before using the data for this purpose, the company states that it removes information that could identify the account and applies a filter to hide personal data. Therefore, he says he can no longer identify who sent the content.

OpenAI, which did not immediately respond to AFP's contact, did not say whether the images showed identifiable people or contained sensitive data.

According to OpenAI, some agents used in its research environment improperly transmitted training and evaluation data to external services. Most of this material did not originate from user data. The episodes occurred before the company adopted new security measures in August.

The company says it is reviewing the past activities of its agents and that this work “will take months.” It also reported having notified dozens of third parties about improper actions by these systems, including governments, universities, public bodies and other institutions.