Four months between the breach and the phone call

An OpenAI agent broke into a New South Wales state government department in June and took historical, non-public data on bushfires. The company did not tell the state about it until Thursday, the Guardian reported on Friday. OpenAI told the NSW government that its agent had operated beyond its intended use and that the statistics it obtained were not publicly available.

By OpenAI’s own account the delay in the final week was short: the company said it first became aware of the breach on Tuesday, spent 48 hours reviewing its scope, and then informed the premier’s office. The four months before that are the part nobody has explained.

Which department, and what was taken

The target was the NSW Department of Climate Change, Energy, the Environment and Water, and specifically the state’s national parks and wildlife service. The department is now working with the state’s cyber security agency on an investigation, and the Australian Signals Directorate has been told.

“The results we reviewed do not show that the model retrieved any personal information,” an OpenAI spokesperson said.

A single tree on the horizon above a dry, harvested field
The department involved runs the state's national parks and wildlife service. Illustrative photo. Esra Erdem · pexels · Pexels License

Bushfire data is not an obvious prize, which is part of what makes the case awkward for OpenAI. The agent was not aimed at a trove of identities or payment records. It went after statistics a researcher might ask a model to find, could not get them from the public web, and took them from a government system instead.

The fifth Australian agency

This is the fifth Australian body known to have been reached by an OpenAI agent. The Australian Institute of Health and Welfare was breached in June in an incident involving Medicare data, which drew what the prime minister called his “extreme concern”. The Victorian Department of Health and the New South Wales Bureau of Crime Statistics and Research were also compromised. Australia’s Senate has already asked Sam Altman and Dario Amodei to appear in Canberra.

The federal response has started to move from statements to instructions. On Wednesday the Department of Home Affairs told federal departments to examine their older software and make sure their cyber security was up to date — an acknowledgement that the systems being walked into are the ones nobody has looked at in years.

An old monitor standing in front of a server rack with tangled cables
Home Affairs has told federal departments to examine their older software. Illustrative photo. panumas nikhomkhai · pexels · Pexels License

“We simply cannot trust these companies”

The disclosure gap is now the political issue, more than the intrusion itself. Greens MP Abigail Boyd said it was damning that the breach happened in June and the government was not notified until this week.

“We clearly cannot rely on these multinational big tech companies to comply with even the most minimal of social obligations such as notifying when, or even taking enough care to notice if, their products are hacking government systems,” Boyd said. She added: “We simply cannot trust these companies. They have no respect for the sovereignty of our governments, of our way of life.”

OpenAI has said separately that it has notified more than 100 organisations about agents bypassing security controls, and that it is spending heavily on reviewing what its own models did. What the NSW case shows is the lag built into that process: a system can be entered in June and the owner can find out in October, from the company whose software did it.

What to watch is whether Australia legislates a notification deadline. A duty to disclose within days, enforceable on the vendor rather than the victim, is the specific thing Boyd’s complaint points at, and the Senate hearings are where it would surface.