The Australian government said an AI agent from OpenAI gained unauthorized access to the government's health data portal in June of this year. This may be the first known case of an artificial intelligence independently hacking a government website.
The Australian government said an AI agent from OpenAI gained unauthorized access to the government's health data portal in June of this year. This may be the first known case of an artificial intelligence independently hacking a government website.
As Australian Prime Minister Anthony Albanese said during a speech at the UN General Assembly in New York, an AI agent gained unauthorized access to the Medicare health statistics database while conducting a study of government health care spending, Reuters writes .
"According to available data, there was no large-scale compromise of the entire network. However, this situation is completely unacceptable," Albaniz stressed.
The Prime Minister expressed disappointment with OpenAI's delay in informing the authorities: the government was only informed of the leak, which occurred in June, on September 10. The Australian authorities have also expressed "extreme concern" to OpenAI CEO Sam Altman and are reviewing three more government medical sites that may have been compromised.
In its official statement, OpenAI noted that an internal review did not reveal any access to patients’ personal health records. The company explained the incident as a failure during information retrieval: “We recorded activity on several Australian government sites and services when our models attempted to find answers to queries… The models took actions that we did not plan.”
The case has strained relations between Canberra and US tech giants. Australia has previously drawn criticism from the US and social media for introducing the world's first social media ban for children under 16, as well as new rules to turn off algorithmic news feeds.
The Australian government has now created a special working group to investigate the incident and assess the reliability of protecting government networks from similar AI actions.
