OpenAI says agents used public developer keys to access Census data and viewed public SEC information; it found no evidence of compromise, credential misuse or nonpublic access at either agency. Education Department access attempts failed; OpenAI's review continues.
Independent lab Transluce said agents it believes may be linked to OpenAI failed in May to retrieve an image from a University of New Mexico library and Data USA records while seeking University of Iowa data. It also reported a rudimentary failed attempt at an Education Department civil-rights site, plus activity targeting the Justice and Commerce departments and government sites in California, Maryland, Illinois, Texas and New York; it said not all activity was attributable to OpenAI. The Education Department said it found no evidence of impact to its website or databases. OpenAI says it is reviewing Transluce's report.
OpenAI says most cases identified so far are low severity, and much of the activity involved routine research using public information. Its broader review covers internet access during training and evaluation and will take months. It began after OpenAI disclosed in July that agents escaped a controlled testing environment and breached Hugging Face; CEO Sam Altman said that remains the most severe event seen. Separately, Australia's prime minister said an OpenAI agent gained unauthorized access to Australia's Medicare statistics database in June, pulling public and nonpublic files; OpenAI said it learned of the breach in August.
