They called it "misaligned model activity", when perhaps it's better to just call it what it is - "hacking".
The language around the hacking by an AI agent into an Australian Medicare data portal earlier this year is the clearest example yet of why Australia needs to get serious about its approach to appropriate safeguards for AI.
If alarm bells haven't been chiming loud enough to wake us all up, the chillingly bureaucratic and unapologetic statement regarding an email that landed in a mid-level public inbox at Services Australia two weeks ago should have them fairly clanging by now.
We now know that an OpenAI agent - that is a clever but fake thing engineered by clever but real people - gained unauthorised access to the Medicare statistics reporting service portal administered by Services Australia, all the way back on June 18.
According to Prime Minister Anthony Albanese, who revealed the breach to the public on Thursday, the AI agent accessed both public and non-public files, but it doesn't appear anyone's personal Medicare details were accessed.
Mr Albanese, currently in New York, also detailed the "frank" discussion he'd gone on to have with OpenAI's chief executive Sam Altman, about the length of time it had taken to notify the Australian government, and the fact this had been done via a public inbox.
It's this last detail that would seem to indicate Australia's standing in the eyes of the tech lords, even though Mr Albanese maintained Mr Altman had "clearly accepted that the company had not done good enough".
It's dismissive and cavalier, but also slightly passive-aggressive, to send what amounts to a "due diligence" email to an address that may well have been the first to pop up on the "Contact Us" page of Medicare's main website.
But in fact, the revelations can and should be a major turning point in Australia's relationship with AI providers. The breach is concerning, but not particularly surprising, given we've already heard of AI agents going rogue and attacking each other.
We've even had a former researcher from Anthropic go viral with a warning on social media that AI could kill us by the end of the decade. He did so in a resignation letter. A more senior Anthropic colleague subsequently confirmed this belief among insiders.
Meanwhile, here in Australia, a taskforce has been set up to review Australia's AI cyber security preparedness. The so-called "rapid" review will consider possible law enforcement and legislative responses and how to ensure such an incident is not repeated.
It is clearly too late for such a taskforce, as the horse may well have bolted on the cyber security front.
The opposition has been quick to rightly point this out, with leader Angus Taylor accusing the Prime Minister of failing to preempt such a breach. And ACT independent senator David Pocock has also called out the government's lackadaisical approach to what is a growing and now quite well-known existential threat.
"We have these companies warning of existential threat and hacking critical government infrastructure and the Labor government's response is 'We'll get to that next year'," he said in a statement.
Indeed, at time of writing, there is yet to be consensus as to whether OpenAI has in fact breached any Australian laws, hence the taskforce.
Who knows how compromised Australian security will be by the time we see any "rapid report"?
.png?w=600)
