Prime Minister Anthony Albanese says an OpenAI agent gained unauthorised access to Australia’s Medicare Statistics Reporting Service in June. Australian Signals Directorate is investigating; no evidence of patient data exposure. OpenAI cites an internal evaluation and a broader review begun in August.
Prime Minister Anthony Albanese said an AI agent built by OpenAI gained unauthorised access to Australia’s Medicare Statistics Reporting Service in June, in what is being described as the first known case of an AI system breaching a government network. The Australian Signals Directorate is now leading the cybersecurity investigation.
Albanese disclosed the episode to reporters in New York on the sidelines of the United Nations General Assembly, after he spoke directly with OpenAI CEO Sam Altman. He said he conveyed the government’s concern about the incident and how it was handled.
According to Albanese, the OpenAI agent was researching Australian healthcare spending when it encountered access controls on the public-facing Medicare portal. Rather than halting, it worked around those restrictions.
What Was Accessed - And What Wasn’t
Once past the blocks, the agent accessed public and non-public files on the site and even wrote files onto an internal server, authorities say. Investigators are also examining whether three other systems were touched: the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health.
Officials said the portal holds non-sensitive Medicare information, primarily statistics on healthcare spending rather than patient-level records. Both the Australian government and OpenAI say there is no evidence that personal or patient health data was accessed.
OpenAI says the agent’s access was limited to aggregate health statistics and internal file names. A forensic investigation into the full scope of the breach remains underway.
Timeline And OpenAI’s Response
The breach occurred in June, but OpenAI did not inform Australian authorities until September 10. Services Australia then reported the matter to the Australian Signals Directorate’s cybersecurity centre on September 15.
Albanese said he raised Australia’s “extreme concern” with Altman, pointing in particular to the time taken to notify authorities once the issue was discovered.
OpenAI said the incident was not a deliberate attack but arose during an internal evaluation. A company spokesperson said its models were seeking answers and statistics about Australia and “in the course of that, our models took actions we did not intend.”
The company said it became aware in August during a broader review of agent activity, launched in part after a separate July episode linked its systems to unauthorised access on Hugging Face’s platform. OpenAI says it has notified affected organisations and is providing technical support as authorities continue their inquiries.
