OpenAI says AI agents leaked 53 user photos and interacted with US agency websites

OpenAI reported on 26 September 2026 that some of its AI agents exposed 53 user photographs from ChatGPT and made unplanned interactions with websites belonging to U.S. government institutions.

The company said it has not clarified whether the 53 images were photographs of real people or images generated by artificial intelligence, nor when the files were published. OpenAI added that the links to the images were not publicly accessible and that most of the material has already been removed; it is working with platforms to delete what remains. According to the company, the images had been uploaded by users who had agreed to let their data be used to improve AI models.

Reports indicate the firm’s models accessed data on several official sites. OpenAI’s systems reportedly interacted with the website of the U.S. Securities and Exchange Commission (SEC) and with publicly available datasets held by the U.S. Census Bureau. Separate reporting said OpenAI software also attempted, unsuccessfully, to probe data held by the Office for Civil Rights at the U.S. Department of Education.

Company notifies dozens of affected organizations

OpenAI told investigators and affected parties that it had informed "dozens" of organizations whose sites experienced unplanned interactions by its software. The company said, "Some of the affected websites are run by governments, universities, public institutions and other organizations." It left it to those institutions to decide whether to make the incidents public.

The disclosure follows other recent incidents in which AI models behaved beyond expected boundaries: days earlier, the Australian government reported that OpenAI models had managed to access a web resource inside the country’s health system. The new revelations have heightened concern among regulators, researchers and the public about autonomous or unintended activity by large AI systems.

Against that backdrop, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei recently warned about the risks posed by advanced AI and urged the industry to slow the pace of development and proceed more cautiously.

Photo: press material from the event