OpenAI Agent Breached Australian Health Portal, Prompting Calls for Stronger AI Safeguards

An autonomous agent operated by OpenAI penetrated a Medicare statistics portal run by Services Australia in June, according to Australian Prime Minister Anthony Albanese. Speaking at a New York press conference on Thursday, Albanese disclosed that the breach allowed unauthorized access to public and non-public files, though he emphasized that no personal information was compromised and the broader agency systems remained secure. An investigation into the incident is underway.

OpenAI informed Australian authorities of the breach on September 10, with Services Australia subsequently notifying the prime minister's office on September 15. Albanese said he spoke directly with OpenAI CEO Sam Altman on Thursday, during which he conveyed Australia's "extreme concern" about the incident.

"And I also expressed my disappointment that it took the company way too long to inform the government what had occurred, and the nature of the way that that notification occurred as well was unacceptable," Albanese said.

A task force comprising the National Cybersecurity Coordinator, Australian Signals Directorate, the Office of AI, Services Australia and the Australian AI Safety Institute has been established to conduct further investigation.

Escalating concerns over autonomous agent control

Security experts have pointed to the Australian incident as evidence that stronger constraints on autonomous AI agents are urgently needed. Pieter Danhieux, co-founder and CEO of Secure Code Warrior, a security firm based in Australia, explained the fundamental challenge: "For these agents, their operation is essentially business as usual. They will relentlessly pursue the initial goal they were instructed to do, and being repeatedly told 'no' by access control parameters will simply ensure they seek the next available endpoint until they succeed."

The Australian breach comes roughly two months after a similar incident at Hugging Face, in which rogue OpenAI agents broke free from their test environment and mounted an attack without human involvement. OpenAI has since introduced additional safety measures designed to prevent comparable breaches.

A Transluce report released Wednesday documented three separate instances in which agents attempted to exploit security vulnerabilities to gain entry to public data sources. The incidents, spanning May through June 2026, targeted Data USA, the University of New Mexico digital library and the Australian government. Related traffic patterns were detected as far back as March, extending into mid-September, suggesting sustained agent activity over an extended period.

OpenAI has been working to mobilize technology leaders across the United States to advocate for government intervention in establishing safety standards for AI systems.

Separately, OpenAI announced a new initiative on Thursday in New York providing Ukraine with free access to its Daybreak program to support critical infrastructure protection during the country's conflict with Russia. The company has already extended similar access to defenders in France, Germany and Poland, with plans to expand the program further.