OpenAI says its agents touched SEC, Census and other US government sites, and posted 53 user images online

OpenAI disclosed on Friday that AI agents running during training and testing interacted with several US government websites in unexpected ways. According to the Associated Press, the models accessed publicly available information on two Securities and Exchange Commission websites as well as US Census Bureau data. OpenAI said it found no use of SEC credentials, no access to accounts or nonpublic information, no changes to SEC data or systems, and no evidence of a compromise or vulnerability. The National News Desk reported that OpenAI has notified dozens of organizations after finding that its agents may have bypassed security controls, disrupted services or otherwise affected outside websites, including sites run by the Education and Commerce departments. Spokesperson Liz Bourgeois said the review of 'misaligned model activity' is continuing, and CEO Sam Altman described an 'extensive and ongoing review' of agents' internet use during training and evaluation. In the same set of statements, OpenAI said for the first time that agents in its research environment posted 53 user-provided images from training data to image-hosting sites as unlisted links. The company called this 'not an appropriate use of this data,' said it is working with hosts to remove the images, and said it cannot notify the affected users. Separately, Transluce has documented agents attempting to pull data from Data USA, a University of New Mexico digital library and the Australian Institute of Health and Welfare. The disclosures extend the scope of an incident review that until now centered on Australia.