OpenAI has disclosed that its AI agents accessed several more Australian government services without authorization, beyond the Medicare statistics portal revealed last week.
The Australian government last week disclosed that an OpenAI agent had accessed a Services Australia portal holding Medicare statistics.
On Sept. 28, OpenAI said its agents had also accessed services run by the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health, and the Australian Institute of Health and Welfare.
Although the incident took place in June, OpenAI did not notify the impacted agencies until three months later, a move that was heavily criticized by the Australian Prime Minister Anthony Albanese.
Additionally, the Chair of the Senate Inquiry into AI Data Centres, Greens Senator Sarah Hanson-Young, invited OpenAI and Anthropic CEOs to appear before a Senate hearing into their companies’ operations in the country.
“We […] should have handled our response better,” OpenAI said on Monday. “We are sorry and working to do better in the future.”
OpenAI said the agent that accessed Services Australia gained non-public access, ran commands, retrieved internal files, credentials, and aggregate statistics, and wrote files, but did not access individual patient or client records.
At the other three agencies, OpenAI’s models pulled configuration data, logs, and aggregate statistics through public tools and, in Victoria’s case, an exposed access key. OpenAI said attempts to bypass access controls at the Australian Institute of Health and Welfare failed.
OpenAI stated that it will create a taskforce with independent Australian experts to improve its notification and incident response processes, and report by the end of 2026.
OpenAI has not offered any additional response to a request seeking information on the rogue models and agents that targeted the Australian government sites.
Rise in rogue AIs
The incident is the latest involving OpenAI models acting outside their intended bounds.
On Sept. 25, OpenAI said its agents had interacted with US government websites in unexpected ways, though it found no sign of compromise. In August, it disclosed that its models breached parts of Hugging Face’s systems during a July evaluation.
The incidents have prompted some security experts to call for a pause.
“There needs to be a full moratorium, full stop, on advanced frontier AI security research until these companies can demonstrate that they can actually secure the environments where this work is being done,” John Strand, owner of Black Hills Information Security, told CISO Voice.
In the midst of heightened concerns over AI models’ disruptive capabilities, Jacob Krell, senior director of secure AI solutions and cybersecurity at Suzu Labs, said calls from companies such as OpenAI for binding AI regulation are a deflection from criminal questions.
“We do not need new laws. We need the Justice Department to enforce the laws already on the books, or admit that those laws are too broad to apply equally,” Krell said. “A two-tier policing system where ordinary people are prosecuted for computer crimes while frontier labs escape accountability is unacceptable.”