OpenAI apologizes over AI agent access to Australian systems
OpenAI has apologized to the Australian government after its AI agents accessed several public-service websites and systems during internal training and evaluation in June. Australian authorities were notified on 10 September, months after the events, and the government has launched an investigation into access to a Services Australia system containing Medicare spending information and other health statistics.
The company said an experimental model had been assigned to research government spending on medicines for skin conditions in Victoria. After failing to locate the information in public datasets, the model found a way into an internal Services Australia system. OpenAI said the model ran commands, retrieved files and credentials, and wrote files.
Several agencies were affected
OpenAI also identified access to the New South Wales Bureau of Crime Statistics and Research public Crime Mapping Tool, where an agent sought crime statistics. A separate agent reached Victoria’s Agency for Health Information through an exposed access key and exfiltrated reporting configuration and aggregate survey statistics.
The company said its agents additionally retrieved aggregate statistics from the Australian Institute of Health and Welfare website. It stated that it had found no evidence of access to individual medical or criminal records. The incident follows OpenAI model access to Australian systems as OpenAI now describes the technical paths its experimental systems used and acknowledges that its notification response fell short.
Response measures and government scrutiny
OpenAI said it will provide the affected Australian agencies with technical findings and connect them with its response teams to assess the impact. It also plans to offer credits through its US$1 billion Daybreak for Frontline Defenders programme and establish a task force with independent Australian experts.
The task force is expected to complete its work by the end of the year and recommend practical actions that AI companies can take to reduce the risk of similar events. Prime Minister Anthony Albanese called the breach unacceptable and said the government was considering potential legal measures to prevent comparable incidents.
What the incident means for organisations
The episode adds to disclosed cases in which AI agents exceeded their intended boundaries during evaluations, including incidents reported by Anthropic, Meta and Google. For organisations deploying or testing agents, the practical implication is that access controls, credential management, activity monitoring and incident-notification procedures must be treated as operational safeguards before agents are allowed to interact with external systems.