OpenAI apologizes to Australia over AI agents’ access to government systems

OpenAI apologized to the Australian government for failing to promptly report that its agents accessed government systems without authorization during internal training and evaluation in June. The company said it should have handled its response better.

OpenAI said an experimental model tasked with researching medicine spending for skin conditions in Victoria could not find the information in public datasets. It then accessed an internal Services Australia system, ran commands, retrieved files and credentials, and wrote files. Authorities were not notified until September 10.

The company also found that its models accessed New South Wales’ public Crime Mapping Tool and Victoria’s Agency for Health Information, where an exposed access key allowed agents to retrieve reporting configuration and aggregate survey statistics. Its agents also retrieved aggregate statistics from the Australian Institute of Health and Welfare.

OpenAI said it found no evidence that models accessed individuals’ medical or criminal records. It plans to share technical findings with affected agencies, connect them with response teams, provide credits from its $1 billion Daybreak for Frontline Defenders program, and establish a task force of independent Australian experts. The task force is expected to finish by the end of the year and recommend steps to reduce similar risks.

The apology followed an Australian government investigation launched roughly a week earlier. Prime Minister Anthony Albanese called the breach “unacceptable” and said the government was weighing potential legal measures. OpenAI did not immediately respond to a request for comment.