Rogue OpenAI agent targeted Australian government site

Australian Prime Minister Anthony Albanese said an OpenAI agent hacked into a government health agency in June, gaining unauthorized access to both public and non-public files. 

Albanese, speaking at a New York press conference Thursday, said the breach involved data from a Medicare statistics portal administered by Services Australia. No personal information was accessed, he said, and there was no broader compromise of the agency. Authorities are conducting a further investigation. 

OpenAI notified Australian authorities about the incident on Sept. 10, and by Sept. 15, Services Australia notified the prime minister’s office. 

Albanese said he spoke to OpenAI CEO Sam Altman on Thursday, expressing the country’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 added.  

Albanese said a task force, which includes the National Cybersecurity Coordinator, Australian Signals Directorate, the Office of AI, Services Australia and the Australian AI Safety Institute, would look further into the matter.

Additional guardrails

Cybersecurity leaders said the Australia breach shows that additional measures are necessary to limit the behavior of autonomous AI. 

“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,” said Pieter Danhieux, co-founder and CEO of Secure Code Warrior, a security firm founded in Australia.

The breach is being disclosed two months after the Hugging Face incident, where rogue OpenAI agents escaped their test environment to launch an attack without any human authorization. OpenAI has since developed additional safety measures to prevent such an attack in the future. 

A report by Transluce on Wednesday outlined three separate incidents where agents attempted to break into public data sources through the exploitation of security vulnerabilities. The incidents, which included attacks on Data USA, the University of New Mexico digital library and the Australian government, took place between May and June of 2026. 

The report showed related traffic dating back to March and continuing into mid-September, an indication of how active the agent activity has been. 

OpenAI has rallied technology leaders across the U.S. to push for additional government action to help develop standards designed to increase safety. 

OpenAI announced a program Thursday in New York to provide Ukraine free access to its Daybreak program, in order to help support critical infrastructure in the country amid its ongoing war with Russia. The company has provided access to defenders in other countries, including France, Germany and Poland, and said it plans additional work to extend access in the future.