September 25, 2026, 8:01 am | Read time: 2 minutes
An autonomous AI agent from OpenAI gained unauthorized access to a government portal in Australia. The incident has sparked political criticism and raises new questions about the control of powerful AI systems.
AI Bypassed Access Restrictions
An incident in Australia highlights the unexpected paths modern AI systems can take. An autonomous AI agent from the U.S. company OpenAI accessed a portal of the state health insurance Medicare, even though the desired information was initially not provided there.
Australia’s Prime Minister Anthony Albanese called the incident “unacceptable.” According to him, the AI accessed both public and non-public files of the portal. The system was originally tasked with researching medical and health-related statistics. According to authorities, the agent initially navigated several government websites like a regular user.
Also of interest: OpenAI Reveals New Problem Cases
Now Even a Gym Is Hacked by AI
“Extremely Unlikely That OpenAI Wanted This”
Government Criticizes Delayed Information from OpenAI
Problems arose only with the Medicare portal. When the requested data was not released, the AI apparently found an alternative way to bypass the existing restrictions. According to Albanese, the agent did not accept the rejection and independently sought a way to obtain the information.
Additional criticism is directed at the communication from the AI company. According to the Australian government, the incident occurred back in June. However, authorities were not informed until September. Albanese stated on the sidelines of the UN General Assembly in New York that he had conveyed his “utmost concern” to OpenAI CEO Sam Altman. He also made it clear that the late notification of the government was unacceptable. The information was merely sent via email to a public inbox.
What the AI Actually Saw Is Still Unclear
As of now, no personal data of citizens has been accessed. According to current knowledge, individuals are not affected. The affected portal mainly contains information on expenditures and other statistics related to Australia’s state health insurance, according to the government.
However, it is still unclear what data the AI actually accessed and to what extent information was retrieved. A forensic investigation is now set to determine what exactly happened and whether other government systems might be affected. OpenAI stated that an internal review found activities of its models on several Australian government sites. The systems attempted to find answers and available statistics about Australia. According to the company, the models performed actions that were not originally intended.