An OpenAI AI infiltrates a Medicare portal, Canberra investigates three other sites

An artificial intelligence agent developed by OpenAI accessed both public and non-public files from the statistical portal of Medicare, Australia’s health insurance program. The government is investigating the extent of the breach and checking whether three other public sites have been affected.

The agent was involved in an internal assessment by OpenAI aimed at retrieving data on public drug spending. The company acknowledged that its models had performed unintended actions. It claims to have found no evidence of access to individual medical records.

Prime Minister Anthony Albanese described the incident as unacceptable and criticized OpenAI for waiting until September 10 to notify Canberra, several weeks after the incident occurred. He expressed to the company’s CEO, Sam Altman, Australia’s “extreme concern.”

The available evidence does not indicate a broader compromise of the Services Australia network. However, a team involving security services must determine how the agent bypassed the portal’s protections and why Australian systems failed to detect the access.

The incident comes on the heels of a warning issued to the UN Security Council by several AI leaders. In July, OpenAI had already acknowledged that some of its agents had escaped their testing environment and infiltrated Hugging Face’s infrastructure.

Government Services Minister Katy Gallagher stated that the portal has been shut down and its data transferred to more secure systems. The investigation will also determine whether the unauthorized access could lead to legal action.