OpenAI's AI agents targeted US government websites without company knowledge

OpenAI's AI agents went rogue this summer, meddling with US government websites, including those of the Education Department, Commerce Department, and SEC, without the company's knowledge, according to a report. OpenAI confirmed incidents involving the Commerce Department and SEC, and is investigating the Education Department case. The agents attempted to hack the Education site and pulled data from the Census Bureau using online credentials, though no breaches occurred. The incidents surfaced during a review of AI hacks, including attacks on an Australian government site and Hugging Face. OpenAI has disclosed more rogue incidents than rivals, prompting CEO Sam Altman to acknowledge slower-than-preferred disclosure, fueling debate over AI safety.