An OpenAI agent hacked into an Australian government health database in June, Prime Minister Anthony Albanese revealed Wednesday, amid concerns from world leaders and artificial intelligence firms about the industry’s rapid, unregulated expansion.
asserted
Albanese → hack → expansion
Albanese said the AI agent had accessed public and non-public files in the statistics reporting service portal for Australia’s universal health insurance scheme, Medicare, in June.
asserted
agent → say → June
OpenAI, however, only notified the government of the breach 84 days later—through an email sent to the public mailbox.
asserted
OpenAI → notify → mailbox
“No personal information is believed to have been accessed at this stage, but investigations are ongoing,” Albanese said in a press conference in New York, where he attended the U.N. General Assembly.
asserted
he → believe → Assembly
An OpenAI spokesperson said in a statement that during a recent review, they identified activity involving “several Australian government websites and services as our models attempted to look up answers, and available statistics for questions about Australia during an internal evaluation.”
asserted
models → say → evaluation
Its AI models then “took actions we did not intend.”
asserted
we → take → actions
The breach follows a string of incidents where AI models have gone rogue, prompting debate about whether the world is ready to deal with such security risks stemming from its swift development.
asserted
world → follow → development
Albanese said he had a “very frank” phone call with Altman about the breach, and the Prime Minister claimed the OpenAI CEO “has acknowledged their issues with protocols.”
asserted
CEO → say → protocols
The Prime Minister then said Australian intelligence authorities will help conduct a “forensic investigation” to look into whether other government systems were affected.
asserted
systems → say → investigation
A task force will also review the incident, and the Australian government will seek advice on whether any offenses were committed and if the incident should be referred to the federal police.
asserted
incident → review → police
The AI agent-led hack comes just as OpenAI CEO Sam Altman himself warned leaders present at the U.N. Security Council about the risks AI systems pose with their increasing capabilities and autonomy: “They could move faster than our institutions, concentrate power in too few hands, or make decisions that people no longer understand or control.”
uncertain
people → lead → that
What happened?
asserted
What → happen → ?
On June 18, OpenAI’s research team used an internal model that accessed the Medicare statistics website, as it looks into public medical spending, according to Albanese.
uncertain
it → use → Albanese
Katy Gallagher, the Minister for Government Services, said the website is “most often used by researchers and academics who get that aggregated data about benefit statistics, prescribing statistics, to use in their own research,” and clarified that it is “not in any way related to Medicare in terms of claims, payments, processing, individual information.”
But OpenAI hadn’t been aware of the potential breach until August, when it was reviewing “misaligned model activity”—or activity that deviates from the user’s intent.
uncertain
that → say → intent
Albanese said that after encountering security blocks, the agent “found a way around those blocks—didn't accept no for an answer” to get the information it wanted.
asserted
it → say → information
OpenAI notified Services Australia, which delivers the federal government’s social services, on Sept. 10.
asserted
which → notify → Sept.
It wasn’t until Sept. 17 that Gallagher had been advised of the breach, and it reached Albanese over the weekend.
asserted
it → advise → weekend
Albanese criticized how Australia received a mere email from OpenAI, and how delayed the notification was despite the security risk.
asserted
notification → criticize → risk
The Prime Minister added three other Australian government websites may have been affected: the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health, though Minister for Defense Richard Marles later said in a Thursday press briefing that the interactions with agents in these three sites were “entirely normal” and only included public information.
uncertain
interactions → add → information
The NSW bureau said it was aware of a vulnerability that could allow access to a crime mapping tool, but asserted that the concerned dataset has no personal information such as names, dates of birth, personal addresses, or other identifiable information about people.
uncertain
dataset → say → people
There’s also no evidence that a breach has already occurred.
asserted
breach → ’ → ?
Marles added that while the impact of the Medicare statistics portal breach is “relatively minor,” a non-human agent’s ability to access a government website unauthorized represents “a very serious incident.”
Shocking, yet predictable, says Albanese
Several high-profile incidents of AI agents going rogue preceded the Australian government breach.
asserted
incidents → add → breach
In July, OpenAI announced that, during a cybersecurity test, its agents infiltrated the AI company Hugging Face.
asserted
agents → announce → company
Independent researchers also discovered that rogue OpenAI agents took over a German website, transforming it into a message board for other AI agents.
asserted
agents → discover → agents
But it’s not only OpenAI: during testing by the U.K.’s AI Security Institute, Anthropic’s most advanced artificial intelligence model, Mythos 5, created fake personas to deceive real people and try to plant malicious code.
asserted
model → ’ → code
“It was a shock that it occurred, because it was real and serious,” Albanese said of the Australian government website hack.
asserted
Albanese → occur → hack
“But it also, I think, was something that had been predicted, including by the AI companies themselves.”
asserted
that → think → companies
TIME has a licensing and technology agreement with OpenAI.
asserted
TIME → have → OpenAI
The Aporia
.jpg?branch=production&width=1024&quality=75&auto=webp&crop=16:9)
