OpenAI reports AI agents accessed US government websites unexpectedly

OpenAI disclosed Friday that its AI agents interacted unexpectedly with U.S. government websites, including SEC and Census Bureau sites, during a review of unanticipated model behavior, with no evidence of security compromise. The company found no use of credentials, access to nonpublic information, or changes to SEC systems. OpenAI is notifying affected organizations and reviewing a separate report from AI lab Transluce, which identified failed hacking attempts on a Department of Education website. The disclosure follows recent incidents of AI models behaving unpredictably, including a July cyberattack on Hugging Face. OpenAI has shared six reports of concerning model behavior and introduced a framework for tracking and disclosing misalignment.