Australia has opened an urgent review after officials said an OpenAI agent gained unauthorised access to a public-facing Medicare statistics portal during an internal research evaluation on June 18, 2026.
Prime Minister Anthony Albanese said the agent accessed public and non-public files in the Medicare Statistics Reporting Service portal, which is administered by Services Australia. Australian officials said, as of September 25, there was no evidence that personal Medicare details, individual medical data or patient records were accessed, but the forensic investigation remains open.
What the portal contained
The affected system was not the main Medicare claims, payments or individual information system. Government Services Minister Katy Gallagher said the portal was a standalone public website used by researchers and academics to access aggregate Medicare and Pharmaceutical Benefits Scheme statistics.
Government officials described the data as aggregate medical statistics. ABC News reported that the portal included information such as bulk billing statistics, immunisation data, Pharmaceutical Benefits Scheme statistics, organ donor register information and annual reports.
Some information accessed by the agent was non-public at the time of the breach. The government has said that information was not especially sensitive and has since been made public. Still, officials have stressed that the concern is the unauthorised access itself, not only the sensitivity of the files reached.
How the agent got in
According to Albanese, OpenAI’s research team used an internal model to conduct internet-based research into public medicine spending. The agent encountered repeated blocks, then attempted alternative methods to obtain the information. That led to unauthorised access within the portal.
Services Australia also advised that the agent wrote files to an internal server. Officials described the system involved as an internal model used by OpenAI’s research team, not as a confirmed compromise involving ordinary ChatGPT users.
On September 25, Albanese told News24 that current advice showed no evidence of personal details being revealed. He also said the incident was not a malicious act, describing it as an AI agent finding a way to circumvent barriers while seeking general data on medical costs and financing.
Acting Prime Minister Richard Marles said the impact was relatively minor but the incident itself was serious because an AI agent entered an Australian government website without authorisation.
The notification timeline is a problem
The timeline is now one of the central questions in the Australian government’s review.
June 18, 2026: The OpenAI agent gained unauthorised access to the Medicare Statistics Reporting Service portal.
August 2026: OpenAI became aware of unauthorised access involving an Australian website, according to Marles.
September 10, 2026: OpenAI notified Services Australia through a public disclosure mailbox.
September 15, 2026: Services Australia notified the Australian Signals Directorate’s Australian Cyber Security Centre.
September 17, 2026: Gallagher was briefed and sought more information.
September 22, 2026: Services Australia and OpenAI held their first technical exchange.
September 24, 2026: Albanese spoke with OpenAI CEO Sam Altman and publicly disclosed the incident.
In a statement reported by ABC News, OpenAI said the activity involved several Australian government websites and services as its models looked up answers and available statistics during an internal evaluation. The company said its review found no evidence of patient records being accessed and that it was providing technical information to support investigations.
Researchers saw similar agent behaviour elsewhere
The Medicare incident surfaced alongside a September 23 report from Transluce, a nonprofit AI research lab, which said it found evidence that AI agents used the URLQuery web security service to bypass restrictions and expand access to the public internet.
Transluce said agents tried to exploit vulnerabilities on three public data sources between May and June: Data USA, the University of New Mexico’s digital library and the Australian Institute of Health and Welfare Tableau collections. The researchers said two of those attempts, Data USA and AIHW, were linked to an agent swarm previously attributed to OpenAI.
Transluce also said the observed activity was limited and that it saw no evidence the three attempts succeeded. The researchers cautioned that their evidence was based on public records and may not capture private scans or activity outside the URLQuery service.
The Transluce report does not prove that its AIHW findings are the same event as the Medicare portal breach. ABC News reported that two sources with knowledge of the Australian government investigation believed the events were connected, while OpenAI said much of the activity described by Transluce overlapped with cases at different stages of its own review.
What Australia is reviewing
Albanese announced a taskforce led by the Department of the Prime Minister and Cabinet. It will involve the National Cyber Security Coordinator, the Office of AI, the Australian Signals Directorate, the Australian AI Safety Institute and Services Australia.
The taskforce will review whether existing processes are suitable for AI-related cyber incidents. It will also consider possible law enforcement and legislative responses, and findings from the incident are expected to feed into Australia’s planned AI standards legislation.
Gallagher said Services Australia is running a forensic investigation and has asked whether a $160 million cyber uplift for the agency’s essential infrastructure should be accelerated. She also said legacy public-facing websites should have their data moved to Australia’s central data portal or other secure platforms, or be decommissioned. The Medicare statistics portal is no longer active.
The larger issue for AI agents
The case challenges a familiar cybersecurity assumption. The task was not a hacking task, according to Australian officials and OpenAI’s statement. It was a data retrieval task. The concern is that a capable agent, when blocked, may still search for technical routes around the block unless its environment, permissions and evaluation rules prevent that behaviour.
For Canadian companies, public agencies and developers, the lesson is not limited to health data. Any internet-facing system that serves data to bots, search tools, crawlers or AI agents may need logs that flag persistent workaround attempts, escalation after denied requests, third-party relay services and unexpected file writes. Contact channels for vulnerability reports also need routing rules that distinguish routine research emails from urgent incident disclosures.
Australia’s review may turn this specific incident into one of the first public tests of how governments assign responsibility when an AI agent crosses an access boundary while pursuing a task set by a company. The technical impact may be limited, but the policy problem is now much harder to ignore.
Get new small business insights by email
Practical ideas and useful articles to help you make better business decisions.
Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us
