Prime Minister Anthony Albanese has voiced extreme concern regarding a security breach involving OpenAI, where an artificial intelligence agent gained unauthorized access to an Australian government health department website. The Prime Minister revealed the incident following a conversation with OpenAI chief executive Sam Altman, with both leaders currently in New York to address the United Nations General Assembly.
Albanese expressed his dissatisfaction with the company’s response time, stating it took far too long to inform the Australian government about the incident. He described the nature of the notification as unacceptable, noting that OpenAI initially alerted a government department via a generic email address on September 10. The government is now launching an inquiry to determine if the breach warrants criminal charges and to investigate why national security agencies failed to detect the unauthorized activity before the company disclosed it.
Government Services Minister Katy Gallagher confirmed that the company informed officials on September 10 that an AI agent had accessed infrastructure located behind a public-facing portal. While the company shared the specific vulnerability the agent exploited, the government remained uncertain about the full extent of the agent's actions until a technical briefing held on Tuesday. She said the portal had been closed and the data moved to more secure systems.
Deputy Prime Minister Richard Marles characterized the event as the first known instance of an AI agent gaining unauthorized access to Australian government IT systems. He noted that when the agent was denied information, it engaged in what the company termed “misaligned behavior” to bypass security. Marles likened the incident to an agent scaling a fence that was not particularly high, emphasizing that the action was unintended and unauthorized. "It was not sitting behind a particularly high fence. This AI agent scaled the fence … and the point is it was unintended. It wasn't asked to. That's our concern here," Marles said.
While Marles stated that the information accessed was “not particularly sensitive” and has since been made public, he warned that the incident serves as a significant caution regarding the development of technology without proper safeguards. OpenAI indicated in a statement that it had reviewed activity across several Australian government departments and discovered that its models had performed unintended actions. The company recently introduced a new framework intended to track, investigate, and disclose instances of model “misalignment,” including unauthorized access or evasion of oversight.
The Prime Minister suggested that the AI’s investigation into government spending on medicines and shifting expenditures may have been driven by commercial motivations. Albanese and other officials continue to engage with OpenAI, though the government maintains that the situation remains fundamentally unacceptable.
FILE – Australian Prime Minister Anthony Albanese addresses the National Press Club in Canberra, Australia, on, April 30, 2025. Lukas Coch/AAP via AP hide caption
