A rogue Open AI agent hacked “non-public files” in Australia’s universal health insurance agency in a breach labelled an “extreme concern” by Prime Minister Anthony Albanese.

It’s understood to be the world’s first instance of an autonomous AI agent breaching a government website to independently access private data.

The US-based artificial intelligence firm confirmed the unauthorised access to Service Australia’s data portal on June 18 after an AI agent acted independently while conducting health research.

The agent also interacted with sites belonging to the Victorian Department of Health, NSW Bureau of Crime Statistics and the Australian Institute of Health and Welfare.

Mr Albanese has personally called OpenAI boss Sam Altman to confront him over the breach by its AI agent which acted independently while conducting health research.

“Today, I spoke with the CEO of OpenAI, Sam Altman, to express Australia’s extreme concern about this incident,” he said.

Mr Albanese added that he also expressed “my disappointment” that it took the company “way too long” to inform Australia of the breach.

The tech company has since issued a statement admitting its models “took actions we did not intend”.

Mr Albanese said while it was understood AI agents didn’t access personal information, he confirmed “investigations are ongoing”.

The company claimed they only found out about the June 18 breach in August before then contacting the Australian government through a public facing general email address on September 10.

The email address was called public.disclosure@servicesaustralia.gov.au which is a dedicated contact to report cyber security incidents to the department.

Senior Labor Senator Katy Gallagher claimed Services Australia checked the inbox the following day on September 11 and picked up the email.

It was only on until September 15 that Services Australia identified the email and escalated the seriousness of it to the government, including by alerting the Australian Signals Directorate.

Opposition leader Angus Taylor questioned the timeline of the announcement and why Australia hadn’t identified the suspicious behaviour when it occurred.

“We need more detail on those timelines as to when the government was informed of this,” Mr Taylor said.

Just a day earlier on September 14, OpenAI’s VP of global policy Ann O’Leary had been in Canberra to attend an Australian Strategic Policy Institute event on AI where she met with senior Australian officials.

The government has started the process to establish a taskforce and has sought “urgent advice” on whether the breach should be referred to the Australian Federal Police.

The PM declared the incident “obviously unacceptable” at a press conference in New York on Wednesday local time during his week-long trip to the UN General Assembly.

“This incident occurred in June this year, and involved an OpenAI agent gaining unauthorised access into the public-facing Medicare Statistics Reporting Service portal, which is administered by Services Australia,” Mr Albanese said.

“Evidence currently available (shows) there is no broader compromise to the Services Australia network. Nonetheless, this situation is obviously unacceptable.”

Defence Minister Richard Marles echoed his concern on home soil, labelling it “deeply unacceptable” but insisted while it was a “very serious” incident the impact was “minor”.

“No individual’s information was accessed. There’s nothing in the site that has been compromised. So, whilst this is a very serious incident, the impact actually is relatively minor,” Mr Marles told News24.

Mr Marles later described the behaviour by the AI agents around the Medicare data as “misaligned” behaviour at a Sydney press conference and reiterated that the data wasn’t personal.

He also referred to the interactions with the other government websites as “normal”.

At the same press conference, Senator Gallagher said she found out on Thursday, September 17 and other ministers were informed over the weekend.

Senator Gallagher on Thursday sought to further clarify the exact content the AI agent accessed.

She explained that the AI agent had been undertaking a task by OpenAI to “conduct internet-based research into public medicine spending” as part of an “internal capability evaluation”.

Senator Gallagher said it had jumped the fence to gain access to “publicly available aggregate Medicare and Pharmaceutical Benefits Scheme statistical data”.

“Services Australia was notified by OpenAI that an AI agent had accessed infrastructure behind the public-facing Medicare Statistics Reporting Service portal,

“The Medicare Statistics Reporting Service portal is a standalone website.

“It’s public facing, and it is not in any way related to Medicare in terms of claims, payments and processing individual information.”

“It’s a completely different system, and the two shouldn’t be conflated.

“It’s most often used by researchers and academics who get that aggregated data about benefit statistics, prescribing statistics to use in their own research.”

Communications Minister Anika Wells said the incident showed why Australia’s moves to regulate tech companies, including through the Digital Duty of Care, was so important.

“Overnight’s example clearly demonstrates that big tech does not feel beholden, in any way, to make themselves accountable in Australia despite literally using their data,” Ms Wells told reporters in Brisbane on Thursday.

The government has now established a taskforce led by the Department of the Prime Minister and Cabinet to investigate the incident.

It will work with the Australian Signals Directorate and AI Safety Institute to examine what the AI agent did, how it gained access and the extent of the impact.

Mr Marles said the incident also highlighted the need for stronger safeguards as artificial intelligence rapidly develops.

“I mean, I think you make a valid point and the bigger picture here which is concerning is the need for there to be guardrails and safety measures put in place as this technology is developed.”

He said the technology offered enormous benefits but safeguards needed to keep pace with its rapidly expanding capabilities.

“We saw enormous benefits of humanity but it has to be developed with safety and what that means is that the guardrails, the safety measures need to be well ahead of the capability that is being developed and that’s actually what needs to occur here.”

Mr Marles said Australia was working cooperatively with OpenAI as the taskforce investigates the incident, but reiterated the government’s anger over what occurred.

“You know, fundamentally, this is a very unacceptable situation.”