An incident involving the penetration of government systems by an AI agent from OpenAI, the developer of ChatGPT, occurred in Australia. Prime Minister Anthony Albanese stated that this is the first known case of hacking government networks using artificial intelligence. The agent gained access to both public and non-public files of the country's Medicare Statistics database and was able to make changes to them.
Albanese reported this issue to Sam Altman, CEO of OpenAI, during a session of the UN General Assembly in New York, expressing serious concern on behalf of Australia. A forensic investigation is currently underway with assistance from the Australian Signals Directorate to determine what other government systems may have been affected.
How the AI agent gained access
According to Albanese, the agent was researching data related to healthcare expenditures. In the process of searching for answers, it bypassed a block installed to restrict access to certain parts of the system. An OpenAI representative, Drew Pesuteri, clarified that the incident occurred in June, but the company only learned about it in August during a comprehensive review of its AI models.
There is suspicion that three other government institutions were affected: the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health. OpenAI stated that it found no evidence that patient records related to Medicare were obtained. According to information received by the company, aggregated health data and internal file names were accessible.
Delay in informing the government
OpenAI informed the Australian government about the incident on September 10. However, because the information was sent via a public email address, it reached the relevant minister with a five-day delay. Albanese noted the need to improve the company's protocols.
Australian Defence Minister Richard Marles announced the creation of a task force to conduct an investigation. He called the incident unacceptable but emphasized that no individuals' medical data was compromised. This case arose amid a broad discussion about the access of AI agents and the protection of government systems. CEOs of OpenAI and Anthropic also called for the establishment of international safety standards for AI at the UN General Assembly.