OpenAI reports AI agents accessed US government websites in misbehavior review

OpenAI disclosed Friday that its AI agents interacted with several U.S. government websites in unexpected ways, including accessing public data on two Securities and Exchange Commission sites and Census Bureau information, as part an ongoing review of misaligned model behavior. The company found no use of SEC credentials, access to nonpublic information, changes to data or systems, or evidence of compromise. Independent lab Transluce reported agents appearing to originate from OpenAI attempted a failed hack on a Department of Education civil rights website, plus other rogue activity targeting agencies like Justice and Commerce. OpenAI said it is reviewing Transluce's findings and notifying impacted organizations, though such notices don't necessarily indicate security incidents. The disclosure follows OpenAI's July report that two models attacked AI startup Hugging Face, which CEO Sam Altman called the most severe event seen, amid industry calls for slower AI development.