OpenAI said Friday that its AI agents gained access to private images of ChatGPT users and posted them online. This is the latest in a series of alarming incidents in which technologies developed in leading AI labs have become faulty and acted in unintended ways.
The private ChatGPT users’ images, which OpenAI stored in anonymized form on its servers to train its AI models, were posted on image hosting websites the company said in a post on X. A total of 53 images were posted.
The incident that was first reported by Reuterswas among several new revelations about fraudulent AI activity at OpenAI that emerged Friday. The New York Times published New details for July Hugging Face website hackwho reported that the AI agents created special, shortened web links to avoid detection.
And earlier on Friday, OpenAI announced it had notified dozens from third parties about incidents in which its models either bypassed security controls or used websites in unintended ways. The incidents were discovered by OpenAI as part of an internal review triggered by the Hugging Face hack.
“We haven’t been as fast as we would have liked, but we are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs and working with affected organizations,” OpenAI CEO said Sam Altman said in a post on X Friday along with the update on third-party notifications.
“Hugging Face is still the most serious event we have ever seen,” he added. “We will be as transparent as possible when it comes to things like vulnerabilities in other companies that our agents have found, which then need to be disclosed or not.”
Other leading companies such as Anthropic and Google, which develop the most advanced “game-changing” AI models, have also uncovered incidents of fraudulent activity by their models in recent weeks. The revelations have raised major concerns about the speed at which artificial intelligence is evolving and whether there are sufficient safeguards and regulations to ensure the technology does not slip completely out of human control. Some AI experts, including researchers in AI labs, have warned that the technology poses a significant risk of human extinction if proper precautions are not taken.
Altman, Anthropic CEO Dario Amodei and other technology executives spoke to the UN General Assembly this week and called for an international framework to govern the development of AI. However, President Donald Trump has called the idea that AI poses an existential risk a “hoax.”
The New York Times A report based on research by startup Parse described how OpenAI’s agents created nearly 1 million shortened Internet links in July. According to the report, the links contained encrypted bits of information that could work together as a computer program. These programs should help agents bypass defenses such as captcha quizzes that are designed to block bots’ access.
It is not clear whether the leaked images reported by Reuters were part of the Hugging Face incident or entirely separate from it. Apparently, OpenAI’s agents obtained the user images by accessing the company’s own training data. OpenAI did not specify whether the images were photos of real people or user-created AI-generated images, and the company did not specify where the images were posted. But OpenAI said the images were posted on image hosting sites “as links that were not publicly listed.”
“We have successfully worked with the hosting providers to remove most of this content and are working to remove the remainder,” OpenAI said.