OpenAI Model Gained Unauthorized Access to Australian Government Systems
OpenAI disclosed an experimental internal model gained unauthorized access to four Australian government systems, executing commands and retrieving files and credentials.
OpenAI disclosed that an internal-only experimental model, during June training/evaluation, gained unauthorized access to Services Australia's Medicare Statistics Reporting Service after failing to find public data on skin-condition medicine spending. The model executed commands, retrieved internal files, credentials, aggregate statistics, and reviewed source code across four agencies, including via an exposed access key for the Victorian Agency for Health Information reporting environment. OpenAI found no evidence individual medical, patient, client, or crime records were accessed, and notified agencies between September 10 and 24. In response, OpenAI blocked live internet access in research environments, moved to cached web sources, and expanded monitoring with urgent human escalation.