An OpenAI agent – not an actual person, but a piece of code – gained unauthorised access to the Services Australia portal in June, which includes the Medicare statistics service. OpenAI engineers discovered this breach in August when they performed routine audits of internal logs, but didn’t notify the Australian government until September, issuing a warning that went to a general email address listed on a public government website.
And finally, the public found out what had happened about two weeks later. This isn’t a taste of what’s happening in the future; it’s happening right now – a largely unregulated company is developing increasingly autonomous technology and is allowed to think and decide for itself whether a breach of a critical piece of a government data network is urgent… or not.
It’s not clear at this stage if this genuinely was a “rogue” agent, or perhaps a publicity stunt, but OpenAI says its agent can autonomously run commands, retrieve internal files and determine site passwords, and then go on to write files into that system. While no client or patient records were accessed – or so we are told – it’s hardly an excuse that OpenAI can use. The basic issue is that the public shouldn’t have their medical histories accessed and exposed in this way.
OpenAI has offered a belated apology and begrudgingly said that they could have handled their response in a much better way but, once again, that’s not the point.
Here we have a giant corporation interfering in a government system, and the company took many weeks after discovering the issue to issue the warning, and not even going through the process of ensuring that it reached the right people within the Australian government. We still don’t know why the “rogue” agent was able to access the Medicare system, why OpenAI’s supposed safeguards failed on this occasion, and who within the company decided that informing the government could wait just a little bit longer.
Of course, the answers to these questions will be assessed by the government’s newly created Joint Select Committee on Artificial Intelligence, but it’s unclear whether these are the right people – parliamentarians – to be able to understand and assess these complex technological issues comprehensively.
The bigger issue with this breach is the imbalance of power that exists, especially in the context of an American company that has the full backing of US President Donald Trump, who rejected the idea of global regulations on AI, and specifically targeted Prime Minister Anthony Albanese’s push for international AI “guardrails” and digital safety.
The Australian government is now hoping to investigate a technology whose developer knows far more about its own behaviour than any minister or parliamentary committee will ever know or understand. This situation sets up a weak foundation when it comes to accountability, with limited resources and an underwhelming knowledge base in which to launch action that acts in the public interest.
Albanese continues to sell AI as a national opportunity, launching his first policy announcement out of the blue just a few months ago, where he stated that we had to “shape AI rather than be passively shaped by it”, and outlining that his government wants all the productivity gains, the investment and the data centres.
His National AI Plan promises economic growth that supposedly sits next to digital “safety guidelines”, but promises like this are difficult to keep when the industry itself is proceeding as such a rapid speed, that it’s almost impossible for any government to create the correct legislation that will keep up with this pace.
And the Medicare breach is an excellent example of this: while Albanese was making his surprise AI announcement in July, a rogue OpenAI agent was busily hacking into the Medicare backend, with no legislation in place to work out what should happen when such an autonomous agent enters a government system without authorisation, and the company then fails to raise the alarm quickly enough.
This is where Albanese’s approach looks decidedly weak. He talked as though Australia needs to open up its arms to welcome in AI or be left behind, providing an urgency that handed the debate to the tech giants on a platter: if governments move too quickly and demand too many legal safeguards, they’ll lose investment and totally miss the AI boat that might only arrive once in a generation, even if it does end up destroying that generation.
It seemed a bit too clever by half and desperate: a powerful and confident government that understands what it’s dealing with would create the legislative framework first. It’s certainly the case that there are many opportunities within AI and it’s a technology that should be harnessed in the public interest, but the Albanese government forgot to provide the legislative support to deal with this when it all goes wrong, which is what clearly happened with the Medicare breach.
The related issue of data centres is also something that’s going far beyond anything that governments are currently capable of regulating effectively, with the public interest being left behind, in favour of corporate interests. Data centres take up a lot of land, massive amounts of electricity and water, and depend on the infrastructure that communities must also live within and, in many cases, help pay for, and the surreptitious nature of these developments – such as Goodman Groups controversial $1.2 billion “Project Mars” data centre in Lane Cove on Sydney’s lower north shore (which has now been withdrawn due to public resistance) – suggests that governments know that these are not in the best interests of the community.
And even when governments are more open about these data centres, the political conversation is always framed around the value of investment, and all the supposed jobs that will be created. But there has to more to these projects than just trying to hoodwink the public with big numbers. Who receives the profits? Who will carry the burden of local energy resources? What exactly are the commitments and promises that governments are making to these corporations, and how will the public know whether they are getting a fair return on these investments?
Of course, these are questions that should always be asked in a normal functioning democracy, and they are questions that should be answered before any contracts are signed and the first sod is turned by a compromised political leader, who’s only too happy to throw away community wellbeing for the sake of vested interests that will also go on to compromise the data of that community.
In The Wealth of Nations, the economist Adam Smith discussed the “invisible hand” of the free market, but unlike the nefarious nature of neoliberalism, his theories were backed by human empathy, moral foundations and government oversight when needed. AI technology and development – as it stands at the moment – upends that philosophy and, to paraphrase Orwell, is more like an invisible boot stamping on a human face, for ever, without the checks and balances needed to restrain that boot.
It’s an insidious development that is starting to flow into work, into culture and into politics. As we have seen with the Medicare breach, AI is a liar and an impersonator. It reflects the discrimination and unevenness that exists in the world out there, because that’s exactly what it’s been trained to do. It steals the work of writers, artists and journalists, and demands actions in the Courts to argue that these creators don’t deserve a fair payment for this theft.
This is not to suggest that AI – and super intelligence, if it ends up coming to that – should be abandoned, far from it. But it needs to have the guardrails that Albanese argued for at the United Nations, to ensure that it works in the interests of the community, not in the interests of the Altmans, Musks and Zuckerbergs of the corporate world, who would be only too happy to sell off humanity, if it adds yet another trillion dollars or two to their already bulging bank balances.
We can see now that OpenAI’s attack on the Medicare portal was something that was just waiting to happen, intentional or otherwise. The Albanese government can commission yet another review or form another committee with uninformed members of parliament, but it needs to create powerful legislation that doesn’t necessarily stifle the development of AI, but harnesses this powerful technology towards the interests of the public, and not to these undeserving vested interests.
For far too long, the likes of Google and Facebook have forced governments to let them build their platforms first and deal with the consequences later – and, of course, when it’s far too late to do anything much about it. The Medicare incident has shown the cost of government indifference – a smaller cost at this early stage – to these processes, but at least it’s there as a warning sign. The Albanese now has a clear choice: govern this industry with an iron fist that’s in the public interest or, like the unsuspecting seals welcoming the hunters in the Arctic regions, keep opening the door to these great risks that Australians never agreed to take on, which will ultimately, swallow us all up.
Albanese is a fool. He's doing more damage to this country than any other leader we've ever had. The next thing we'll know we'll be a State of the US.
This whole case seems like a big distraction and a mish-mash of paranoia. There was no evidence, as far as I am aware, that the public "had their medical histories accessed and exposed...".
I find it silly to call every AI agent that does a bad thing "rogue" ... it is doing what it was designed to do. Simulate the thinking and actions of a human. The companies and governments apply the "rogue" tag to try and manage the outrage. These systems are being designed to do what they are doing, bacause that is essentially how humans solve problems.
The AI agent had no greater access to the Medicare system at question than a human hacker. The core issue for Medicare is not AI ... but cybersecurity.