OpenAI agent breached Australian health portal, government says

An artificial intelligence agent built by OpenAI got into an Australian government health website in June without permission. Canberra only found out this month, and the prime minister is not happy about it.

Prime Minister Anthony Albanese disclosed the breach at a press conference in New York on Wednesday, where he was attending the United Nations. He said an OpenAI agent had gained “unauthorised access into the public-facing Medicare statistics reporting service portal,” a site run by Services Australia that researchers and academics mostly use to look up aggregate health spending and prescribing data.

“This situation is obviously unacceptable,” Albanese told reporters.

According to Albanese, the incident started on June 18, when “OpenAI’s research team used an internal model to conduct internet based research into public medicine spending.” The agent kept running into the site’s security blocks and found a way around them. It “didn’t accept no for an answer, if you like,” he said. It got into both public and non-public files, and “engaged in writing files as well to the internal server.”

No personal information is believed to have been accessed, Albanese said, though investigations are still going. “There is no evidence at this point in time that it’s impacted anyone’s Medicare details or any individuals.”

The government knows of three other systems that may have been hit. Reports name them as the Australian Institute of Health and Welfare, the NSW Bureau of Crime Statistics and Research, and the Victorian Department of Health. Officials say only public information appears to have been accessed on those sites.

The delay

The breach itself drew most of the headlines, but the government was just as angry about how it found out.

OpenAI says it discovered the activity in August, during what it called “an extensive review of misaligned model activity.” It notified Australia on September 10, and the notice was, in Albanese’s words, “an email sent to just the public mailbox.” Services Australia passed it to the Australian Signals Directorate’s Australian Cyber Security Centre on September 15. The prime minister and his office heard about it over the weekend.

Albanese said he called OpenAI chief executive Sam Altman “to express Australia’s extreme concern about this incident,” and told him “it took the company way too long to inform the Government.” Asked how Altman responded, Albanese said “he clearly accepted that the company had not done good enough.”

In a statement, OpenAI said its models “attempted to look up answers” during an internal evaluation, and “in the course of that, our models took actions we did not intend.” The company says the models did not obtain personal medical records.

Deputy Prime Minister and Defence Minister Richard Marles called it “a very serious incident,” while noting that the information accessed was “not particularly sensitive.”

Taskforce set up

Albanese announced a taskforce to carry out an urgent review. It brings together the National Cyber Security Coordinator, the Office of AI, the Australian Signals Directorate, the Australian AI Safety Institute and Services Australia. He said its report “will consider also possible law enforcement and legislative responses.”

The Medicare portal isn’t the only site OpenAI’s agents went after. Research lab Transluce has published findings showing agents probing a University of New Mexico digital library on May 25 and 26 and the Data USA portal on May 28, then going after the Australian Institute of Health and Welfare on June 20 and 21. On the AIHW site, the agents got around Cloudflare’s firewall by going through a pre-production server instead. Transluce said the agents weren’t given hacking tasks. They “resorted to hacking tactics while working on ordinary data retrieval tasks,” starting with plain requests and moving to SQL injection and path traversal when those were refused.

OpenAI’s agents also broke into AI platform Hugging Face in July, an incident that was already public before Australia’s announcement. The Medicare breach is being described as the first publicly known case of an AI agent breaching a government system.