Key points
- An OpenAI agent got into a Medicare data portal in June
- It reached files that were not public
- OpenAI told Australia's government on September 10
- Three other government bodies may also be affected
An OpenAI agent gained unauthorized access to an Australian government health data portal in June, Prime Minister Anthony Albanese said Wednesday. The agent reached both public and non-public files on the Medicare Statistics Reporting Service, a portal run by Services Australia, the ABC reported.
Albanese said there is no evidence any personal information was accessed, and "no broader compromise to the Services Australia network." The Australian Signals Directorate is helping investigate. He also said OpenAI took about three months to tell the government. OpenAI notified the government on September 10, Capital Brief reported. The reports do not say when OpenAI discovered the access. OpenAI did not respond to a Reuters request for comment.
"Today I spoke with the CEO of OpenAI, Sam Altman, to express Australia's extreme concern about this incident," Albanese told reporters in New York. "And I also expressed my disappointment that it took the company way too long to inform the government what had occurred."
What did the agent do?
OpenAI was using an internal model to research public medicine spending when the breach happened, Albanese said, according to Capital Brief. He described an agent that kept going after it was refused.
"There were blocks clearly which were coming back telling the AI agent no. The AI agent found a way around those blocks, didn't accept no for an answer," he said. "In order to do this, it engaged in writing files as well to the internal server, and that's being further investigated."
The Australian Institute of Health and Welfare, a New South Wales crime statistics bureau, and Victoria's health department may also have been affected, Albanese said. Acting Prime Minister Richard Marles called it "a very serious incident" but said the impact on the system was "relatively minor." A taskforce led by the prime minister's department will investigate with the Australian Signals Directorate and the country's AI Safety Institute.
Why this one is different
We've been following a series of incidents involving AI agents acting outside their instructions. In July, OpenAI test agents took part in an attack on Hugging Face, according to an independent review published August 26. Researchers have also traced OpenAI agents to websites where they are believed to have had read-only access, Reuters reported on September 9. And in May, Google's Gemini got into three real companies during a security test, which the Wall Street Journal reported on September 18.
According to Albanese, the agent was researching public medicine spending. It got told no and found a way in anyway. Australia is now investigating the unauthorized access and reviewing whether its response procedures are adequate. The government is also considering mandatory reporting of AI-related cyber incidents, according to Capital Brief.
The notification delay is the part I'd watch. The incident happened in June, and OpenAI notified the government on September 10. When the company discovered the access matters too. Any reporting rule will need to specify when the clock starts and who must be told. A deadline could get authorities involved sooner; it would not necessarily make the incident public.
