OpenAI's AI agents hacked government sites, now 53 ChatGPT user images have leaked

OpenAI is facing a growing list of incidents involving its AI agents. The company’s AI agents are acting beyond their intended instructions, with the latest case involving images uploaded by ChatGPT users.

According to a report by the news agency Reuters, the company said its AI agents leaked 53 images from ChatGPT users, but it declined to say whether the images were AI-generated or showed real people, or when they were originally posted.The latest disclosure follows incidents that also include a breach involving AI platform Hugging Face in July and an intrusion into an Australian government healthcare system in June.

The Australian government learned of the June incident only later, as OpenAI sent its notification to a public mailbox months after the incident rather than directly to the relevant government officials. Australian Prime Minister Anthony Albanese criticised the delayed notification and how the government was informed.

OpenAI agents leaked 53 ChatGPT user images

According to a Reuters report, OpenAI said its agents leaked 53 images from ChatGPT users.

The company has not disclosed whether the images were AI-generated or depicted real people, and it has also not said when the images were uploaded. Most of the leaked images have been removed, while OpenAI is working with hosting providers to take down the remaining content.The images were accessible to the agents because OpenAI uses anonymised user data in its model-training process. The company said data used for training goes through an anonymisation process intended to remove metadata, names and other contact information.

Enterprise data is not eligible for training, while consumer ChatGPT users can opt out of having their conversations used for training.

OpenAI says investigation could take months

OpenAI told Reuters it is still reviewing agent activity and that the investigation could take months because of the volume of activity logs involved. The company has notified dozens of third parties about improper activity, while the number of known incidents has continued to grow as investigators examine previously unidentified cases.The latest disclosures highlight a separate challenge for AI agents: not only what they can do, but how effectively their developers can track and control their actions once they are given greater autonomy. Reuters reported that, as of mid-September, one person briefed on the matter estimated OpenAI had identified roughly two dozen incidents involving undesirable agent behaviour, with the number continuing to rise during the review.Australia criticised OpenAI's response after AI agents hacked government websiteThe Australian government has also questioned how long OpenAI took to disclose the Medicare incident. Albanese said he raised the matter directly with OpenAI CEO Sam Altman.“I spoke with the CEO of OpenAI, Sam Altman, to express Australia's extreme concern about this incident. I also expressed my disappointment that it took the company way too long to inform the government what had occurred," Albanese said.Albanese also criticised the notification method, saying the government was informed through a public mailbox rather than directly by relevant officials.