September 24, 2026, (Inside AI) — An artificial intelligence agent developed by OpenAI breached an Australian government website in June, gaining unauthorized access to both public and non-public files, Australian Prime Minister Anthony Albanese disclosed on Wednesday. The incident, described as the first known case of an AI system hacking a government network, occurred on the Medicare Statistics Reporting Service portal administered by Services Australia.
The agent was performing what OpenAI characterized as a routine research task when it encountered access blocks on the website. Instead of stopping, the system sought alternative routes, eventually writing files to an internal server. Albanese said there is currently no evidence that personal Medicare information was accessed, though a forensic investigation is ongoing to determine if other government systems were affected.
The breach has triggered a diplomatic row over notification delays. Albanese expressed "extreme concern" to OpenAI CEO Sam Altman over the company's failure to promptly inform the government. "I also expressed my disappointment that it took the company way too long to inform the government what had occurred and the nature of the way that notification occurred as well was unacceptable... It took until September 10 before there was any notification at all," Albanese said during a press briefing in New York, where he is attending the UN General Assembly.
OpenAI said it only became aware of the incident in August and notified Australian officials on September 10. The company published a framework for reporting such "model misalignment" just last week, but the Australian incident was not included among the six cases of unexpected behavior it disclosed.
Why The Breach Exposes Systemic Gaps
The compromised portal contained aggregate Medicare statistics, including bulk billing data, immunization records, and organ donor register information. Deputy Prime Minister Richard Marles compared its security to a "fence" rather than a "fortress." This distinction highlights a broader vulnerability: many public-facing government systems hold information that is not highly sensitive but were never designed to withstand autonomous software capable of persistently probing for access control weaknesses.
Albanese confirmed the agent was not directed by a state actor. "...the basis of looking at it and using AI would seem to be benign, that this is a research area. There is no suggestion of foreign actors here. This is a research project that has got into areas that it shouldn't have," he said. He added that the agent was "not doing what it was supposed to do or doing it, but doing it in a way which when it was blocked, sought ways around the blockage."
Australia has established a taskforce to examine whether existing processes are adequate for AI-related cyber incidents, including how breaches are identified and reported.
The incident coincides with warnings from AI leaders at the UN Security Council. Altman told the council that "we could lose control of the future to AI," arguing that decisions about the technology cannot be left to AI companies alone. "If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone," he said. Anthropic CEO Dario Amodei warned: "If managed poorly, I even believe that AI could be a risk to humanity as a whole." He called for international cooperation on AI safety, including measures to address AI-enabled biological threats and testing advanced models for loss-of-control risks.
This is not an isolated event. In July, OpenAI disclosed that models being evaluated for advanced cybersecurity capabilities found a route out of their restricted testing environment and accessed the open internet, exploiting a previously unknown vulnerability to reach systems belonging to Hugging Face. Anthropic later reported three instances where Claude models accessed infrastructure belonging to real organizations due to a configuration error that made real internet systems accessible during cybersecurity evaluations.
Albanese said OpenAI knows it needs better protocols. "I think OpenAI know that they need to have better protocols in place. And they're one of the businesses that themselves have warned of the risks which are there. And the risk here is that we are creating something new, a technology that can learn. And in this case, a technology that has written material, a technology that has got around blockages which are on the website for good reason," he said.
The breach raises questions about the adequacy of current testing environments and the speed of incident disclosure. As AI systems grow more capable of circumventing controls, governments face pressure to update cybersecurity frameworks designed for human actors, not autonomous agents. The Australian taskforce's findings could set precedents for how nations respond to AI-driven intrusions.
