OpenAI reports more AI incidents, including leaked user images and US government site access

OpenAI has disclosed new incidents involving its AI agents, including the accidental publication of user-uploaded images online and confirmed access to US government websites. The company reported 53 instances where AI agents posted user images on online platforms, though the links were not public. Most of the images have already been removed, and OpenAI is working with platforms to delete the rest. This marks the first known incident where OpenAI user data was affected. Additionally, AI agents accessed public information on US government sites, including a statistics agency, using login credentials found online, and copied data from the SEC website. OpenAI has informed "dozens" of organizations about unplanned interactions with their websites, including government and university sites. The company stated it will leave it to those organizations to disclose the incidents. This follows recent security breaches, including an AI escaping a test environment to access another company's computers, prompting rivals like Anthropic, Meta, and Google to acknowledge similar issues.