By Noah Secomb, Sarah Spina-Matthews and Will Nicholas in Canberra

Australia will punish OpenAI for letting one of its models go AWOL and hack a Medicare website if legally possible.

The world’s first known breach of a government platform by a rogue AI agent back in June was revealed by Prime Minister Anthony Albanese at a press conference in New York on Thursday, Australian time.

A task force will now scour Australia’s legal options to penalise OpenAI, which will involve exploring how to punish humans and companies for the actions of independent bots.

Environment Minister Murray Watt said that if it was possible, the government would press criminal charges.

“One of the things that they (the task force) will be looking at is whether these matters can be referred to the Australian Federal Police … if it’s not possible, then clearly that indicates that we need to change Australian laws,” he told Seven’s Sunrise program on Friday.

The task force convened for the first time on Thursday and is expected to report in a matter of weeks.

But an autonomous bot hacking into systems while off its master’s leash spelled a legal headache, Digital Economy Minister Andrew Charlton said on Friday.

“An AI agent is not a legal person, and in our law liability for this type of incident always sits with a person or a company,” he told ABC Radio.

“The liability has to be traced back to the intent of a person or a company that created or directed the agent.”

OpenAI discovered the hack in August but kept it under wraps until September 10, when it emailed a notification to a generic Australian public service inbox.

Mr Albanese expressed his disappointment about the delay during a phone call with OpenAI boss Sam Altman on Thursday, with ministers later queuing up to label the company’s conduct unacceptable.

The agent involved in the June breach was researching public medicine funding, interacting normally with three other Australian federal and state government platforms, but scaling a digital “fence” and breaking into the Medicare website.

It had not been directed to hack, an OpenAI spokesperson insisted in a statement released on Thursday.

The incident was a turning point in the debate about how to harness and contain AI, Mr Albanese said.

“This was an AI agent acting independently of its instructions, that is learning by itself,” he said during a fireside chat with former prime minister Kevin Rudd at New York’s Asia Society.

“It’s a bit of a wake-up call about the risks.”

The prime minister rejected assertions he had timed his announcement of the breach to coincide with him and 21 other world leaders issuing a statement calling for more global AI guardrails.

He said he had first heard of the incident after arriving in New York despite the government being notified weeks prior.

“We didn’t want to alarm people unnecessarily … we wanted to make sure that the announcements that were made were fact-based,” he said.

Mr Albanese revealed he had spoken to OpenAI boss Sam Altman to tell him he was going public about the hack as soon as possible.

Nationals Senate leader Bridget McKenzie said Labor had calculated the disclosure was to boost their posturing on AI safety and their international clout.

“There’s undoubtedly been a political decision to delay informing the Australian public,” she told ABC Radio on Friday.

“A decision must have been taken somewhere to wait until the prime minister was on the world stage.”

University of Sydney United States Studies Centre director Olivia Shen agreed the announcement was tactical.

“The prime minister’s over there at the UN talking about the need for global good governance on AI, and we just happen to have had this incident occur that they were notified about last week,” she told AAP on Thursday.

Reporting and reaction times should be much quicker than the three month delay in this case Ms Shen said, adding AI breaches should face similar requirements to cyber security incidents which impact critical infrastructure.

News all day, every day at CityNewsQBN.com.au.