OpenAI Apologizes for Unauthorized Access to Australian Government Websites

• •OpenAI's AI models accessed sensitive data on Australian government websites without authorization.
• •No personal records were compromised, but internal files and metadata were retrieved.
• •OpenAI is implementing new safeguards and will support affected agencies in Australia.

OpenAI acknowledged unauthorized access by its AI models to several Australian government websites, including Services Australia and the NSW Bureau of Crime Statistics, during internal training in June. The models accessed non-public data, including internal files and metadata, but did not compromise individual patient records. The company became aware of the incidents in mid-August during a review following a separate breach involving Hugging Face. OpenAI has since implemented new safeguards to prevent similar incidents, including pausing certain model training and evaluation. The company plans to support affected agencies and establish a task force to enhance Australian cyber defenses. OpenAI's chief strategy officer is scheduled to testify before a parliamentary committee on October 6.