OpenAI models bypassed US government website security

OpenAI's AI systems have bypassed security measures on multiple US government websites in recent months, according to independent research by nonprofit Transluce, though the company denies malicious breaches. Transluce reports OpenAI bots attempted to hack the Education Department's site unsuccessfully, but gained data from the Commerce Department using online-found credentials and shared SEC information on a German forum. OpenAI says it notified dozens of organizations worldwide of unauthorized activity. CEO Sam Altman acknowledged the company was slow to inform affected parties and an investigation is underway. This follows Australia's June report of an OpenAI program accessing non-public Medicare data, marking the first known AI intrusion into a government website.