Thought Leadership
An autonomous AI agent developed by US-based OpenAI hacked a government portal in Australia.
The AI gained unauthorized access to a Medicare portal that provides, among other things, statistics on Australia’s public health insurance system, Prime Minister Anthony Albanese said in New York. The incident was “unacceptable,” the prime minister emphasized.
Speaking on the sidelines of the UN General Assembly, Albanese said he had already spoken with OpenAI CEO Sam Altman and expressed Australia’s “deepest concern” over the incident. He also voiced his disappointment that it took OpenAI “far too long” to notify the Australian government. The incident occurred in June but was not reported to Australian authorities until September, when the company sent an email to a publicly accessible inbox.
What Happened?
Medicare is Australia’s public health insurance system. According to the government, the affected portal operated by Services Australia contains information including spending data and other statistics. The AI agent accessed both public and non-public files. Based on current findings, no personal information was accessed and individuals do not appear to have been affected, Albanese said.
The incident began with what was supposed to be a harmless task: The AI program was instructed to research medical and health statistics. According to Deputy Prime Minister Richard Marles, it navigated several Australian government websites in much the same way as a regular user would.
However, when the Medicare portal did not provide the requested information, the agent found another way to gain access. “The AI agent found a way around the barriers. It essentially didn’t want to take no for an answer,” Albanese said.
Forensic Investigation Underway
It is not yet clear how much data was actually accessed or which files the agent was able to reach. A forensic investigation is now underway to determine exactly what happened and whether other government systems were affected.
OpenAI said an internal evaluation had identified activity on several Australian government websites and services. The models were attempting to find answers and publicly available statistics related to questions about Australia.
“In doing so, our models took actions that we did not intend,” an OpenAI spokesperson said.
(dpa/Translation: Editorial Team)
