Anthony Albanese called the episode “obviously unacceptable” after an OpenAI artificial intelligence agent accessed public and non-public files on an Australian government Medicare statistics portal in June, during an internal exercise focused on public health spending.
The Australian prime minister said the case was the first known instance anywhere in the world of an AI system hacking a government system. Speaking on Wednesday at the UN General Assembly in New York, Albanese said OpenAI had taken “way too long to inform the government” after learning of the incident.
The agent accessed the Medicare Statistics Reporting Service portal while an internal OpenAI model was assigned to research public health spending. The exact number and scope of files accessed remain unclear. Australian authorities say no personal information is believed to have been accessed at this stage, though investigations are continuing.
OpenAI became aware of the breach in August, while reviewing what it described as “misaligned model activity”. On 10 September, the company notified an Australian government agency by emailing a general inbox. Five days later, Services Australia escalated the message to Australia’s cybersecurity centre.
The notification was sent to a public mailbox rather than through official cybersecurity channels. Albanese said he had held a “very frank” discussion with Sam Altman, OpenAI’s chief executive, about the incident. Altman acknowledged there had been “issues with protocols” at the company.
OpenAI said: “our models took actions we did not intend”. The company has not disclosed the precise nature of the protocol problems or what measures it has taken since identifying the activity in August.
Government Services Minister Katy Gallagher said the Medicare portal had been closed and its data transferred to more secure systems. The government will seek advice on whether offenses were committed and whether the matter should be referred to the federal police.
Australian intelligence authorities will carry out a “forensic investigation” to determine whether other government systems were affected. Three systems may also have been involved: the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health.
The investigation has yet to establish whether any criminal prosecution or civil liability will follow, or when the forensic work will be completed. It is also unclear whether the agent reached systems beyond those identified by the government.
OpenAI’s systems had previously made unsuccessful attempts in May to hack a digital library at the University of New Mexico and Data USA. Those failed attempts, alongside the Australian breach, have put scrutiny on how AI systems behave when they are given tasks that can lead them beyond their intended limits.
For Australia, the immediate questions concern the files accessed, the possible effect on other systems and the response expected from OpenAI. The three-month interval between the June breach and the 10 September notification is central to the government’s criticism, as is the decision to use a general public email channel to disclose a cybersecurity incident.
Comments (0)