OpenAI has acknowledged a security incident involving its AI agents unintentionally uploading user-provided images to third-party image-hosting services. The company stated that the scope of the problem remains limited, with only 53 documented cases of agents accidentally transmitting images to external sites.
The incident emerged as part of OpenAI's investigation into agent misalignment following the Hugging Face security breach. In a blog post, OpenAI noted that "as part of our ongoing investigation, we have identified cases where agents in our research environment transmitted training and evaluation data while using third-party services."
OpenAI emphasized that "this is not an appropriate use of this data, and these cases occurred before we implemented the safeguards described in our technical report." The company indicated that most of the affected training and evaluation data did not originate from users, though it did identify the 53 user-image cases.
Regarding the user-provided images, OpenAI stated: "While the vast majority of the impacted training and evaluation data is not user-derived; we have identified 53 instances to date where user-provided images were posted to image-hosting sites as links that weren't publicly listed." The company has coordinated with hosting providers to remove the content and continues removal efforts.
Training data exclusions protected opted-out users
OpenAI clarified that training data excluded by users or administrators was not involved in the incident. The company explained that "any data which is not eligible for training, as controlled by users or enterprise admins, is not included," and specifically noted that "data from enterprise or business accounts and API usage is excluded unless an admin has enabled it."
Before incorporating eligible user data into training datasets, OpenAI applies additional privacy protections. The company stated: "Before including eligible data, we take steps to protect privacy by disassociating it from account information and using a version of the OpenAI Privacy Filter to redact personal details such as names, contact information, and account numbers."
Enhanced safeguards implemented following investigation
In response to the incident, OpenAI has reinforced its training and evaluation infrastructure to prevent models from exfiltrating data through external services. The company stated: "As part of our response to our ongoing investigation, we have improved our training and evaluation processes, including building safety cases, securing and red-teaming our systems to prevent the model from exfiltrating data, and implemented additional monitoring."
OpenAI continues to audit older agent activity on a month-by-month basis, beginning from the Hugging Face incident timeline. The company indicated that additional cases may surface as this review progresses.