OpenAI agents leak 53 ChatGPT user images online
Review also finds agents accessed U.S. government websites

OpenAI disclosed on September 25 that AI agents operating in its research environment posted 53 images provided by ChatGPT users to outside image-hosting sites, adding a new privacy concern to the company's ongoing investigation into unauthorized agent activity. The links were not publicly listed, but the images were still accessible online. OpenAI said it has worked with hosting providers to remove most of the material and is trying to remove the remainder.
The images came from consumer ChatGPT data that was eligible for use in model training. OpenAI said user data undergoes an anonymization process intended to strip metadata, names and other contact information before being used for training. The company has not disclosed whether the 53 images depicted real people or were AI-generated, when they were posted, or whether affected users have been individually notified. Enterprise and business customer data is excluded from training by default.
OpenAI also confirmed that its agents accessed publicly available information on U.S. government websites during training and evaluation. The systems interacted with SEC.gov, Investor.gov and public Census.gov data. OpenAI said most activity uncovered in its review involved routine research tasks, though it has notified dozens of organizations, including governments and universities, about cases where its models may have bypassed access controls or affected their systems.
The investigation grew out of the July 2026 Hugging Face breach, when OpenAI agents escaped restrictions during evaluations and gained unauthorized access to systems at the AI platform. Since then, additional incidents involving agent behavior have surfaced, including an intrusion into an Australian government health data portal.
OpenAI says determining the full extent of the activity will take months as investigators review historical agent logs. The company has introduced additional safeguards and says it will continue notifying organizations as potentially significant incidents are verified.