By The Straits Times | Created at 2026-09-25 21:16:58 | Updated at 2026-09-25 22:54:17
1 hour ago
SYDNEY – Australian Prime Minister Anthony Albanese used a United Nations address on Sept 24 to back an international push to develop artificial intelligence safeguards and “shape AI development rather than be passively shaped by it”.
He delivered his address just as a breach by an OpenAI agent of Australia’s health insurance system exposed a series of gaps in the country’s domestic cyberdefences.
Though Albanese’s Labor government has been a global leader in pushing for social media safety, including imposing a world-first ban on accounts for those aged below 16, analysts say the government has been slow to keep up with the development of AI models.
An OpenAI model on June 18 gained unauthorised access to non-public statistics held by Medicare, the national health insurance scheme.
OpenAI discovered the breach in August, but notified the Australian government only on Sept 10. The breach has led to calls by experts to tighten rules on testing of AI models and on reporting requirements for technology giants.
AI Safety Institute and Office of AI
There have also been calls to boost funding and staff for the government’s two new AI agencies: the AI Safety Institute, which monitors, tests and analyses new AI models, and the Office of AI, which was created in July 2026 to oversee the design of national AI standards.
Rebecca Johnson, an AI evaluation expert at the University of Sydney, told The Straits Times that the Office of AI should help to oversee testing and evaluations of new AI models. She said testing should not only focus on whether AI agents achieve assigned tasks, but also on the methods they use to achieve those tasks.
“Currently, we have no federal oversight of AI models,” she said.
“The Office of AI needs to reach out to the big AI companies and say if they want to operate in Australia, they need to follow our rules and regulations. The companies should provide regular updates on their evaluations, and ensure evaluations are agentic-specific and focus on the pathways that agents take rather than the outcomes.”
The Medicare breach came as Albanese has been pushing for Australia to attract AI investment and work with global technology giants, while ensuring that new models are developed responsibly.
Albanese travelled to California on Sept 18 to meet technology leaders to talk about online safety before travelling to New York for the UN General Assembly (UNGA), where he spoke to OpenAI head Sam Altman about the breach.
Australian Deputy Prime Minister Richard Marles also travelled to California in early September to meet Altman and other technology heads to discuss collaborations with Australia on their cutting-edge AI models.
The Australian government seized on the breach to bolster its case for joining on Sept 21 with more than 20 other countries, including Singapore, in backing a call for control of frontier AI models that was initiated by Finnish President Alexander Stubb and Norwegian Prime Minister Jonas Gahr Store on the sidelines of the UNGA in New York.
Albanese told the UNGA on Sept 24 that the OpenAI breach was unacceptable and highlighted the need to establish standards to ensure that “AI will work and deliver for people – not the other way around”.
“We can’t ignore AI or prevent it,” he said.
He noted that OpenAI gained access to statistics, including public medical spending.
Albanese said on Sept 25 that the OpenAI breach was not “malicious” and had not accessed sensitive personal data.
“This was an act of an AI agent essentially being told ‘no’ (and) when told no, found a way to circumvent some of the barriers that were there,” he told News24.
“There’s no evidence there’s any harm been caused here. But that doesn’t mean that it’s not serious.”
He has launched a task force led by his department to find out whether the OpenAI breach broke any laws, and to investigate any gaps in existing laws and processes covering the government’s use of AI models and AI-related cyberincidents.
The OpenAI breach dominated media headlines in Australia, partly because Medicare is one of the country’s most prized institutions and holds sensitive medical information about most residents.
The government said sensitive data – such as individual medical claims, personal banking details or patient medical histories of Australia’s 27.9 million people – was not accessed.
But the breach has fuelled public concerns about the safety of AI and could make it more difficult for Albanese to press ahead with plans to “make this technology work for us”.
A Lowy Institute opinion poll in June found 64 per cent of Australians believe the risks of AI outweigh the benefits, and only 34 per cent believe the opposite, with the remainder uncommitted.
Faster alerts, government contact and reliable log keeping
Walayat Hussain, director of the AI for Decision Excellence Lab at the Australian Catholic University, told ST that Australia can improve safety by requiring AI firms to notify the government within 72 hours when their models access government systems without permission. He said firms should “keep logs of what the agent accessed and produce an audit trail on request”.
“Australia is not unregulated on AI, but the standards being designed now were for a different problem,” he said.
“The standards the Office of AI is coordinating focus on data centres and copyright. Nothing in them yet addresses an AI agent from an overseas company finding its way into a government system.”
Hussain said the government should set up a contact for technology firms to advise about breaches, noting that OpenAI alerted the government about the Medicare breach by sending an e-mail to a public inbox.
Keeping pace with the changes
He said the government should also conduct staged attacks against its own systems using AI agents as well as human experts.
“The agents get more capable every few months, so a test from last year tells you very little,” he said. “It has to be on a schedule.”
Alastair MacGibbon, former special adviser to the prime minister on cybersecurity, told ST that the Medicare breach was a “wake-up call” for the country.
He said the government needed to significantly boost funding for the AI Safety Institute. The institute has received A$30 million (S$27 million) over four years, compared with about A$120 million annually for an equivalent institute in the United Kingdom.
“The AI Safety Institute needs to be properly funded and start collaborating with labs around the world to test guard rails,” he said.
MacGibbon said Australia had led the world on cybersafety, but now faced the difficult challenge of keeping up with the quickly evolving threats posed by AI.
“Industry and the government have to be better at defending themselves (against AI breaches),” he said.
“Australia is understandably trying to keep pace with a frontier that is moving very fast.”
