Australian agencies warned in May of incoming ‘vulnerability storm’ ahead of OpenAI breach

Por Miriam Webber — POLITICO – TOP Stories

CANBERRA — Australia’s federal government agencies were warned in May of a “vulnerability storm” headed their way as frontier artificial intelligence hit legacy IT systems.

More than 100 agencies were told by Australia’s Home Affairs Department, which oversees the federal government’s cybersecurity policy, that “frontier AI is collapsing exploit timelines from days to hours. Lock down the fundamentals now.”

Home Affairs issued detailed advice May 27 on the accelerating risks posed to “known vulnerabilities, legacy systems and weak cyber hygiene”.

The warning came several weeks before an OpenAI’s agent infiltrated an Australian government website on June 18, accessing Medicare statistics.

“In May 2026, the Department of Home Affairs provided advice to the Chief Security Officers of all 104 non-corporate Commonwealth entities highlighting the risks and benefits posed by frontier AI including actions to manage legacy technology risk,” a department spokesperson told POLITICO.

Home Affairs Secretary Stephanie Foster issued new advice this week following the unauthorized access, directing all agencies to conduct rapid stock takes of their legacy systems, implement targets to reduce the technology and report back on their progress.

The OpenAI breach affected a Services Australia legacy site — an end-of-life IT product — hosting nonsensitive health care data in June 2026.

Government Services Minister Katy Gallagher last week told Services Australia to move data housed on the legacy systems to a newer site, and accelerated access to 160.4 million Australian dollars for “cyber uplift,” initially allocated over four years in the federal budget.

Alastair MacGibbon, former head of the Australian Cyber Security Centre which leads the Australian Government’s efforts to improve cybersecurity, said it would take months if not years for organizations to adapt to the current AI environment.

“One of the realities of IT systems and cybersecurity is it actually takes a long time, and you’re never finished. It’s a bit like painting the [Sydney] Harbour Bridge; you get to the end, and you’ve got to start again.”

“Cybersecurity is an awful lot of hard work, a lot of risk management and planning. AI has upended that,” he said.

MacGibbon said phasing out legacy systems was just one part of effective cyber risk management. “Often the most important systems run on some of the oldest equipment in any organization, whether it’s in the private or the public sector, and that’s because it’s too risky or too costly or too time-consuming to migrate it or update it,” he said.

“So what you do is you’ll put compensating controls in place … the equivalent of airbags.” In these cases officials are deciding that problems are inevitable, but that they can “reduce the harm, or in some cases negate the harm.”

Australian agencies are urged to replace their systems before they reach end-of-life status, but are permitted to run legacy systems as long as they take steps to mitigate risk.

Richard Marles, who is acting home affairs minister, said Australia was acting to “identify vulnerabilities and deal with them before they can be exploited.”

“We have a range of security IT systems across government, and the level of security is relative to the information being protected,” Marles said.

“It is critical that we constantly review those security settings and the systems we have in place.”

OpenAI on Tuesday broke its silence about the unauthorized access, apologizing in a statement and promising to do better.

“One of the ways we intend to take accountability for the situation is to be intentional in working with Australia to help develop practical approaches to how AI developers and governments identify, disclose, and respond to AI cyber behavior, whether malicious or unintentional,” the company said in a statement.

The Australian government has criticized the company for failing to notify it of the incident until September, when it emailed a public inbox that wasn’t monitored regularly.

“We are notifying you of a security vulnerability identified during our review of OpenAl model activity involving Services Australia’s Medicare Statistics service,” the email, reviewed by POLITICO, reads.

“An OpenAl model identified a way to make the server carry out instructions sent through the public reporting interface, without a private account or password.

“It was able to access this to read portions of internal program files and settings, obtain a list of files, and create and read back a small test file on the server.”

The Australian government is also conducting a review into whether its legislative, governance, and information-sharing arrangements are suited to respond to cyber incidents involving AI.

Fonte: POLITICO – TOP Stories