The company "OpenAI" revealed that more than 50 private images of "ChatGPT" users were leaked and published online by the artificial intelligence without obtaining the consent of their owners.
Reuters reported that "OpenAI" stated that the investigation into these incidents, which it described as "security-related," could take months, noting that it had identified 53 instances where images uploaded by users to "ChatGPT" were published on image hosting sites.
It indicated that some of its applications sent data from their internal systems used for training and testing to external sites, including users’ images.
"OpenAI" stated that it has collaborated with hosting service providers to remove most of those images, but it is still working to remove the rest, meaning that some of these images remain publicly accessible.
According to a source familiar with the matter cited by Reuters, "OpenAI" had observed around 24 incidents, as of mid-September, in which the applications behaved "unfavorably."
This is the first publicly announced case where the company’s artificial intelligence applications mishandled user data, and it is the latest example of emerging issues related to uncontrolled behaviors of these systems within "OpenAI."
These images belonged to users who allowed their data to be used in "ChatGPT" for the purpose of training the AI models, as they did not choose to withdraw their consent, meaning they did not request to exempt their data from training.