The AI Medicare wake-up call

0:00This episode of New Politics was released on 2 October, 2026, and produced on the lands of the Wongal and Gadigal people. How did an OpenAI agent hack into Australia's Medicare system? And who is going to make OpenAI accountable for its actions? Big tech keeps telling us to embrace the future on artificial intelligence, but governments also need to put in stronger legislation and safeguards that protects the public interest. I'm Eddie Jockovich. I'm David Lewis. All of this is coming up in this big episode of New Politics. And if you're listening in and haven't subscribed to our website at New Politics, you're missing out on half of what we do. There's regular articles and analysis of this strange world of politics. It's all free and you can subscribe at newpolitics.com.au. And of course if you do want to support us, it's just $5 per month of Patreon and Substack.

1:04And all of this is a good way to support independent journalism. And OpenAI agent has infiltrated an Australian government website. Australian Prime Minister Anthony Albany says an AI agent hacked a statistic portal on the site of the nation's universal healthcare scheme known as Medicare. He said the breach happened in June. OpenAI learned of it in August, but only told the government in early September. The material access is considered non-sensitive, but it adds to the array of concern about the immense power of emerging AI technology and the prospect of it going rogue. Among Australians, it also raises the question of security or lack thereof of government websites. Albany said he called the breach unacceptable and said he's had a frank discussion with OpenAI

2:08boss Sam Altman, adding the company took way too long to alert authorities. Cybersecurity professionals are calling this the first AI agent led government website, hacked, reported anywhere in the world. They say it should raise alarm bells for leaders everywhere. An open AI agent gained unauthorized access to the Medicare statistics portal in June and calling an agent might sound like it's actually a person, but it's not. It's actually an autonomous bot. The company discovered this access through a human inspection of data records and logs in august by their engineers, but waited almost a month to notify the government. That notification then went to a public reporting inbox, which is what anyone can send an email through, which makes this process even more bizarre. But OpenAI has since apologized, which seems to be more like we're only apologising because you want us to apology or sorry, feel that way. But there's a lot of questions about why OpenAI has supposed safeguards failed on this occasion

3:13and why notification to the federal government took so long. Now, the government has said that individual Medicare records were not accessed and making that statement will limit the political damage, but it won't make us forget the failure of this entire incident. And calling this a rogue agent, well, that doesn't give us much confidence at all, David. And I'm not sure if this was actually all accidentally made on purpose or a PR start from OpenAI to say, well, look, this is all the stuff that we can actually do. So be careful. But the end result is that accountability has to remain with the company. It's up to the government to create the legislation that will make them accountable. And at the moment, that legislation isn't there. Yeah, at the moment, we don't quite know who is responsible. Is it the person who pushed a bit harder? Couldn't get the figures they want. So said try something else and the bot does that now that ethically and morally, that may not be the case. But legally, the law does like to find a perpetrator.

4:14Is it Medicare for not having? And they were warned. Medicare was warned that there were certain vulnerabilities in their security. Or is it the fault of the AI company? I think given the nature of the research, I don't think we can really blame the researcher. The research was after general information that really should have been not too hard to get. How much does Australia spend on medicine a year? You might have to collate a few sites and make some assumptions and define medicine clearly. There's medicine include things like panadol and macurecrowman and ointments, those everyday medicines that we have. Or is it prescription medicine? Obviously, how you define it changes the figures significantly. And even if the researcher was an advertising executive or something promoting a new wellness brand, still not an invalid, I still can't say the problem with asking the question. And I think you can fairly argue that it was never the intention

5:16to break into a secure government site. This isn't like in the 1980s and 1990s. When the internet came out and you had those young, brilliant kids breaking into NASA and things like that. And American defense force because we didn't know what security was. I suppose too. It's not fair to blame Medicare for being broken into it. It would be slightly like being charged for having your house broken into. Though to be fair in New South Wales, if you leave your keys in the car or if you don't look your car if it's on the street, you can be fine. Maybe there's an argument that Medicare should have known. I think whichever company it was needs to be regulated and told there are certain things that you can't do, even if you have the ability to do it. And the other thing that I think bears a bit more thought is if somebody with an innocent motive can get in, what about someone with less of an innocent motive? Proper hacker looking to disrupt the system,

6:16either for monetary gain or to break a system. And our bank's vulnerable is our defense forces, vulnerable is our legal system, the courts, computers, all kinds of stuff that is best known public, really. Yeah. Well, we're also making that assumption that open AI is innocent or an innocent operator in all of these. But it's also a question of how the government of Australia or any government for that matter. How can they independently verify open AI as account of this incident when the company controls all of the evidence, all of their evidence, they hold all the evidence and they're based in the United States. As well, and there's virtually nothing the government can do to force open AI to cooperate. So they can have this parliamentary hearing, which has been established. They can have a parliamentary hearing that can ask the questions, but establishing what happened, that requires a forensic investigation by independent experts and technical experts

7:16that have got access to the relevant records and they haven't actually got access to that. You can try forcing open AI to do that, but good luck with that. There's also the question of the dependence of this technology. And as the technology is becoming more embedded within public services and business, governments may end up finding it harder to impose these restrictions. You can see why governments all around the world are wanting to create native AI systems in some sort of a form of AI nationalism. But when you have these promises of safety and rail guards, which is what Elvonezi is promising, the Prime Minister, when you've got that at odds with commercial interests, I think you just really need to have very tight, very strong legislation to force these commercial interests, whether it's open AI or anthropic or anyone else, force them to do the right thing. And that's what good governments should be doing. I note that we've had this ridiculous and really

8:18unenforceable ban on social media for under 16s, which they're claiming is a great success. But I don't see that it is a great success because I can't see the evidence that it is. I still am aware of lots of under 16s with social media accounts. And yet there is seeming reluctance. And look, let's be fair, it takes time to draft the legislation in such a way that it's fairly unassailable in court, that it's fair to all to everybody, not just the users. It's got to be fair to the users. It has to be reasonably fair to the AI generators. It has to make sure that they have a clear and transparent procedure to make sure that these aren't done. That's fine. We live in a system that's supposed to have justice and justice is fairness. We live in a system to where we have innocence until proven guilty. We can't assume under the law that open AI is a automatically guilty when anything goes wrong. In the same way, we can't assume that anyone else's guilty

9:20whenever they are arrested for a crime that the case has to be made. I'm not a fan of tech companies, by the way, I'm not trying to defend them or say, oh, no, leave them alone. It's more a case of the law has to be right. Otherwise, the courts just shoot it down. And no one wants that, especially the courts. The courts are clogged up enough without cases that could have been avoided by a bit more thought. Having said that, I note that some legislation gets done more quickly. This whole notion of data centers, which, on my reading, look like a big con. They create jobs. What jobs? It may be two at the most in the center. And that's not going to knock on into new supermarkets and new police stations and new schools and new business opportunities for people, not for two people. They use a lot of power. They use a lot of water. In a country, too, that we forget at the moment, is one heavy drought away from water shortage. We've had it before here. Or a gambit dam was down to 4% at one point,

10:224% capacity, barren-dong dam, which is seven times the size of Sydney Harbor in the center in New South Wales was down to 2%. You could walk along the bank of the bed of it. So, and with the environment not getting that much better, water shortages are going to hit harder. And suddenly, they want these things that probably aren't needed in the form they're built in. They use a ton of water. So, they need to sit down and look at what are these things allowed to do and what are not allowed to do in terms of security breaches. They may as well look at how do we make them as environmentally and socially and culturally and civically appropriate as possible. Do we need to build so many? Well, therefore, per cent in Warrigamber, dam, let's left over. That's probably enough for the data centers to use up. You know, why not? 4%. That's all good. But the Prime Minister, Anthony Almanese, he did say that he wants Australia to shape artificial intelligence. And that's what he said back in early July.

11:24Roughly at the same time that this open AI was actually interfering in the medicare system. So, there wasn't much shaping going on at that stage. But all of these developments are just moving a lot faster than the legislation can keep up with it. And I don't know exactly what the solution should be, David. Maybe getting AI to develop this legislation that will regulate itself. That's a start. So, you know, it certainly be a lot faster than what the politicians in a committee could come up with. So, with a lot of what Anthony Almanese has said on the protections for the workers of creators and the energy used up by the data centers that you mentioned before, we just need to have some sort of process to work out whether all of this is sustainable or not and whether, well, how much is going to cost as well. Because it pretty much every sector of government or every department of government has always got that question of, oh, how much is all of this going to cost? We need to know the cost of pretty much every department, every government department, except for military

12:25and it seems like AI development at the moment. And the tech industry just keeps saying, oh, look, Australia's going to lose all of this investment unless it starts moving very quickly on this. But corporations are not the ones that should be determining the speed of public policy. And I'm pretty sure that Anthony Almanese is an expert in artificial intelligence. And I'd also suggest that not many politicians are. But the protection of the public is their number one job. That's the number one job of any political leader or any politician. And that's what they have to keep working on. Yeah. Again, I'm not anti-AI as such in terms of, if you're using it to collate information, a bit more sophisticated Google search, that's fine if you're using it for the grunt work of creation, proofreading and layout and stuff like that in music, initial mixes and mastering and things. That's OK. That's just saving a lot of time. It's when people use it to replace things

13:26that can't really be replaced. Human creativity, human intelligence, human strategic. It's like any the horse and cart worked fine, replaced essentially by the motorcar. And turned out there was stuff that motorcar couldn't do that a horse and cart could do. But the motorcar was better at the stuff that it could do than the horse and cart. So again, it's not technology as such. Or it's not AI is all bad all together as such. But it is a tool of limited use. That limited use is very good. But it won't replace the things that people want to replace it with. And when it oversteps its bounds as it has with the Medicare stuff, it needs to stop because, yeah, again, all it takes is someone to save, it got into Medicare essentially accidentally. What can we do deliberately with it? This is New Politics, Independent Analysis of Australian and International Politics. For more free articles and analysis,

14:27join us at newpolitics.com. .au. If society decides that AI is something worth having, there probably is worth having if the correct conditions and legislative controls are set up in the right way that will protect the public interest and make sure that we're getting a fair deal out of it. Out of all of this as a community. And you touched on this before, we have to remember that AI isn't just used for productivity purposes. It's already been used for military purposes in Ukraine and in Gaza with things such as autonomous machines and Robocops which are currently being tested and this field is totally unregulated. So I don't think we should kid ourselves that it's going to be used to make the future of humanity so much brighter like it could.

15:28But there's also these other nefarious ways that it could be used. But I think locally, aside from people using AI and their computers, let's leave the autonomous machines and Robocops discussion for another time. But when it comes to local usage, there is that issue of the data centers and that's a big issue. And that takes up a lot of land, uses up a lot of electricity, a lot of water. And just recently there was a proposal to build a $1.2 billion data center in Sydney's Lane Cove. And for those who are not in Sydney, that's suburban Sydney. That's right in the centre of Sydney. And it's not a remote location. And there was a lot of local opposition to that. And because of that, that proposal has been withdrawn. But this is how this process all works. This one in Lane Cove, that's been knocked back. But then they'll go to another location. They'll try their luck there. They also tried North Cotumber in the Blue Mountains. That's been knocked back as well. But all they need is you get a desperate council

16:30on-side with their promise of jobs that never eventually. You set one up, you get a lucky council on your side. Or you get a compromised council on your side. You set one up there, then you set up another one. And before you know it, there's a data center on every street corner, just like a pub on every street corner. So I think that's where the real problem is. There needs to be a lot of consultation with the community, along with the legislative support to go with it. So you don't end up waking up one morning, and then you see these massive electricity poles installed for a service at the data center, which is exactly what happened in Melbourne a few weeks ago. Yeah. The company's running AI. And this is true of any company. We can't assume that they are working in the public good. They may think they are, but until we can see actual tangible or can be shown actual tangible net positives, we just can't trust them. And that's true. It's true of any large company,

17:30whether it's all worth or, and hello to all our good folk at all, or assisting, or Toyota, or a bank, Westpac, to pick one at random. The individuals working for these companies, maybe, and often are, the very best people you can meet. But the companies are after the one aim, which is to achieve whatever it is they want to achieve. And that's where good people go bad. So the government needs to move with the type of speed that it can move with when it comes to important legislation. I think of the gun laws in 96 that needed an urgent response. The payments in 2008 for the GFC that needed an urgent response. But we need urgent and rigorous action before it spins out of control. I also don't think we're going to end up with Arnold Schwarzenegger walking around looking for John Connor, having been sent back from the past, etc. Or RoboCol, or Judge Dret. Someone pointed out that all of our literature on AI tends to point this way.

18:32So the logical thing for AI as a collater of information is to go that way. So it makes sense. But I think it's less the big change in sentience and all of that. And more small changes that suddenly see a lack of autonomy by those who are supposed to be controlling it. But I think the risks of AI actually goes far beyond that issue of cyber security as important as cyber security is. But there's all of those inherent discriminations. There's a lot of falsehoods that are slipping through AI as well. There's a lot of political bias in there as well, which seems to be very pro-US and pro-Israel and anti-China as well. Now you could argue that this is reflective of the rest of the Western world. There's very Western bias there. There's also very white bias in there as well. But that's where all of these captains of industry come from. And all of these people that are controlling Google or OpenAI, Meta and Fropik,

19:32that's where they come from, these captains of industry. So they will want to perpetuate those biases, whether that's conscious or subconscious, because it's in there interest to actually do all of that. And then, aside from that, there's also the issue of creative content where writers, artists, journalists like ourselves, they also need the consent and payment for their work. And they've done some research into this. And I found out that some of our books have been used to train AI without our permission. And initially I thought, oh, isn't that great? Using small independent publishers to train their system. And then I woke up to myself and realized, no, that's not good at all. Our work has been stolen and we're not being paid for it. And it's just like the rest of society. The markets need limits and regulations. They need common law. These governments have set the rules and regulations for what individuals can do, but not for these large corporations. And I think the example of Google and Facebook or whatever they're called, Alphabet and Meta, I think that they show how platforms can become too big

20:34before the regulations catch up with them. And maybe in these cases, they'll probably never catch up. But I think that the lesson from this Medicare breach, this recent one in Australia, might only be a relatively small incident at this stage compared to what else could have happened. But I think it is the warning sign that Australia needs. And so it can go and sort out a lot of these issues and take back control of its own data and its own information. Imagine what a hostile or a criminally inclined person could do knowing that AI can hack into this stuff. We're forever, of course, getting attempted to know all of services from North Korea, occasionally from China, from Russia, from other foreign governments and other organizations that probably don't have the best interests of Australia at heart. We don't want to make their job easier. We stop most of it, not all, we don't stop all of them because that just happens sometimes.

21:35But we stop most of them. The people in, I think, it's Department of Defense work extremely hard and often extremely effectively. And yet the odd one gets through. And so you get the the optus hacks and the health insurance hacks and things like that. But AI gets smarter, reads our books. And it's the best way for it to get smarter. And it gets to the point where the thing that is hacking becomes more powerful than the thing that is hacked. And suddenly cyber security is meaningless now. It can trace VPNs. So you might think that you're coming from a server based in Belivia or something. But it can work back through to where you're actually coming from. So swift action and sensible action is required. And I believe Australia has the expertise to be able to act swiftly and effectively to fix this. It's just does the government have the will. That's it for this episode of New Politics.

22:37Thanks for listening in. And if you value independent journalism without the influence of corporate backing, join new politics at newpolitics.com.au and support us directly. It's just $5 per month. And we always try to keep it very simple. If you like what we do, please send some support out way. It keeps our commitment to independent journalism ticking along. I'm Eddie Jockovich. Thanks for listening in. And it's goodbye to our listeners. I'm David Lewis. We'll see you next time.