OpenAI identified 53 cases of ChatGPT users’ images being published
Lev Shevtsov26
September
2026
09:54
OpenAI said that its tools posted links to images uploaded by ChatGPT users on third-party websites. As Deutsche Welle reports, the company identified 53 such cases and said it had already removed most of the material.
Links on image-hosting sites
OpenAI explained that artificial intelligence agents sent data to third-party services for training and evaluation in cases when they should not have done so. The company noted that most of the transferred data did not come from users.
The images in question had been uploaded by users of accounts whose owners had allowed their data to be used to improve models. According to OpenAI, the images were separated from the accounts and passed through a privacy filter. Links to them ended up on image-hosting sites but were not displayed in public listings.
More current news is available on the UA.News Telegram channel Telegram.
Review of agents’ actions
OpenAI also confirmed The New York Times report that its tool accessed websites of U.S. federal agencies. The company said that in this case, only publicly available information was obtained.
According to OpenAI, the cases of unauthorized dissemination occurred before the introduction of a new set of protective measures more than a month ago. In July, the company reported that its internal cybersecurity assessments had found that models were bypassing restrictions intended to isolate them from the internet. OpenAI CEO Sam Altman called it the most serious event the company had observed.
Research laboratory Transluce said that agents, which it assessed could have originated from OpenAI, attempted a simple hack of the website of the U.S. Department of Education’s Office for Civil Rights. According to the laboratory’s findings, the attempt was unsuccessful. Transluce also reported other actions against websites of U.S. government bodies and state authorities, not all of which the laboratory directly linked to OpenAI.
