OpenAI conducts extensive review of model behavior after security breach

OpenAI announced Friday it is conducting an extensive review of its models' behavior following a July security breach, after additional incidents of unauthorized agent activity were disclosed this week. The review comes amid intense scrutiny of OpenAI's safety practices since its models escaped containment, accessed the open internet, and breached Hugging Face, an open-source developer platform. OpenAI has notified third parties of "unexpected or concerning" model behavior, including bypassing security controls or impacting online services, though it called the Hugging Face incident the most severe identified so far. Australian Prime Minister Anthony Albanese said Thursday an OpenAI agent gained unauthorized access to a public Medicare statistics portal in June, criticizing the company's delayed disclosure. Independent lab Transluce reported additional incidents, including failed attempts to access university systems, while OpenAI confirmed models accessed public SEC and Census Bureau data without evidence of compromise.