Never miss an episode. Follow The Bloomberg Australia Podcast today.

Australia is embracing the AI boom, with billions of dollars pouring into data centers and the technology promising to transform the economy. But anxiety is growing too. A rogue OpenAI agent’s unauthorized access to Medicare systems has raised fresh questions about what happens when increasingly autonomous AI behaves in ways its creators never intended.

Bloomberg’s Chris Bourke is joined by Asia tech reporter Newley Purnell to unpack what the incident tells us about the risks of AI, whether Australia has the power to keep US tech giants in check, and the dilemma facing Canberra as it pushes for greater safeguards while trying to attract a potentially enormous wave of AI investment.

DisclosureFurther listening: Why the RBA Could Hike Again And for a daily wrap of the business, finance and economic stories that matter to Australians, from Bloomberg's reporters around the globe, sign up to our free Australia Briefing newsletter.

Here is a lightly edited transcript of the conversation: This week, ChatGPT got a grilling in Sydney. Well, its creator did anyway, in a hearing that was held a week after OpenAI was forced to apologize for that rogue AI hack on Medicare.

But how much can Australia afford to push back on the US tech giants while also convincing them to invest billions here? And does Canberra actually even have the power to keep this rapidly evolving technology in check?

To unpack Australia's AI dilemma, I'm joined by Bloomberg's Hong Kong-based tech reporter, Newley Purnell. Newley, welcome to the podcast. Thank you, Chris. I'm happy to join.

So Newley, you watched a few hours of testimony from OpenAI, Anthropic and others at the parliamentary inquiry here on Tuesday. OpenAI flew in their chief strategy officer, Jason Kwon, who again apologized for that Medicare breach. But what did we actually learn that we didn't know before, and was there anything that surprised you?

Right. Well, as you said, it opened with Jason Kwon of OpenAI expressing the company's deep apologies. He said that they could have done better. They were trying to improve their internal processes.

And I think two things jumped out at me as being the most notable from their hearing. One was, you know, there have been questions about what OpenAI knew about these unauthorized agents accessing Australian government websites and when they disclosed it. Of course, we know Prime Minister Albanese has said that they didn't do it fast enough. Some people wondered when OpenAI chief executive Sam Altman met Deputy Prime Minister Richard Marles in California earlier in the month if they had discussed it.

And Jason Kwon was asked directly about this. He said that Sam Altman, in fact, didn't know about it when he met with the Deputy Prime Minister. I don't think that had been said publicly before.

But he definitively said that. And then I think the second thing that jumped out was we got a little bit more detail on how OpenAI says they're improving their safety mechanisms to try to stop some of these actions happening in the future.

So speaking of safety mechanisms, let's go back to the incident that triggered all this. So an OpenAI agent was given a pretty standard task and when it ran into an obstacle, it found a way around that and gained unauthorized access to an Australian government system, in this case Medicare. Look, how worried should we be about that behavior as these agents become more capable and autonomous?

It's troublesome. And it really gets at, I think, the deeper worries about how AI technology is evolving. I think it started earlier with these existential questions, with researchers from some of these companies, for example, resigning and saying that the companies aren't taking these big doomsday threats seriously enough. And that's all in the abstract.

The companies haven't really shared much about their technical systems, the dangers, potential dangers of the technical systems. But this was really a material example of how these things can break out of what's supposed to contain them and actually take action in the real world. So, in other words, we're moving from the theoretical to the actual practical and material. Now, while what they did was not catastrophic — you know, they didn't hack into a nuclear reactor or a utility or commit a cyber crime — they still acted in ways that they shouldn't have.

So I think it really puts actual detail on some of these concerns about how this technology can escape and do things it's not supposed to do.

Yeah, just you mentioning those nuclear reactors kind of fills me with a lot of dread. Prime Minister Anthony Albanese has used that Medicare breach to call for greater standards here on AI. He's talked about establishing Australian standards. Do you know what that could actually mean in practice?

I mean, I guess the bigger question is what powers does Australia currently have to regulate these powerful companies that are based in the US?

Right. Well, he's talked about middle powers, smaller countries like Australia, you know, not just letting these big AI titans get their way, and said that these smaller countries have to work together. It's unclear what a country like Australia could do to them. And one interesting element that emerged in the hearing was a discussion about why OpenAI disclosed it and whether they were legally required to in Australia. And they said it was a voluntary disclosure. And so I think that gets at the idea that, well, they actually weren't legally required by Australia.

So one thing Australia could do, and what some other countries have talked about doing, is creating laws that say if your AI system does something it's not supposed to and affects some other entity, you will legally be required to do that.

And so it seemed to emerge in the hearing yesterday, with some of the lawmakers asking them, well, exactly why did you disclose it? Was it based on your internal standards or legal requirements? And they said it was a voluntary disclosure. And in fact they said, we would welcome countries creating legal regimes that would require us to do this because it would provide clarity.

And one of the things that they said about the delay in when they disclosed it to Australia was that they were trying to work out the facts, but also how significant it was.

And I wonder also if they were determining whether they were legally required to, or whether their own standards said they had to. They said it was our own standards. But I think it gets at the murkiness and, yes, questions over the legality, the legal requirements for companies. So that would be one thing Australia could do, is to make an actual legal requirement that if something like this happens in the future, companies would have to disclose it. Right?

But surely the problem is that this technology seems to be moving much faster than any government can regulate it. We saw OpenAI having to scrap the release of its recent AI model due to safety fears. And there are those, of course, increasing fears around autonomy. Look, does Australia have a realistic chance of even keeping pace with this technology?

I think it's difficult. You know, one of the powers that, say, the US has is that the companies are located in the country, right? And so they have staff there, they have investors there, or they have a lot of users there. And so not just in the US, but any country, when they have actual people there, staff working there, they have a little bit more leverage to be able to, say, summon them to come in and talk to the government.

So the US, you know, has more power in that regard. But it's a question that many countries face, particularly when US or Chinese technologies are being deployed there without staff from the companies there or major investors or huge user groups.

I think on the plus side, Australia, you know, is an advanced economy, consumers have money. And that speaks to the companies' bottom lines.

You know, they care about a country like Australia because there's a decent-sized population and people who are willing to pay for their products. That can't be said of other, say, emerging economies that might have larger populations, but fewer people willing to pay. So oftentimes the bottom line is something that matters to these companies.

And also another reason why they might care about Australia is this massive AI investment boom. We apparently have — Westpac estimates the country's data center pipeline is worth about A$175 billion, roughly A$110 billion of that likely to come over the next three years.

I know you've been reporting on the rise of data centers across Asia. How significant do you think this could be for Australia's economy?

It could be significant, now with the caveat that it could be significant over the longer term, less than the near term. I think a misconception that some people have is that data centers create a lot of jobs. They do create jobs in the construction phase, but oftentimes these facilities don't require a lot of humans to actually run them compared to other, you know, facilities of their size.

But I think over the long term, you know, the case from the AI companies is that these technologies, as they advance, will lead to economic growth and to the real improvement of humans' lives.

So, for example, Jason Kwon from OpenAI gave the example of a survey, he said, in which a number of small businesses in Australia said that they're seeing productivity gains, they're saving time in the jobs they do. He referenced using ChatGPT for medical research.

So the companies say, you know, this stuff will really provide a lot of value over the long term. On the other hand, there was just some research that came out from the US venture capital firm Andreessen Horowitz saying that we know that about half of adults in the US use AI services, but they found that just a tiny proportion actually pay for them.

So that leads to the question of whether these are gonna be newfangled, essentially Google on steroids. Are these gonna be just like fun apps you can do a lot of research with, which is how most people are using them now, or will they really lead to the kind of structural revolution and economic gains, productivity gains?

The companies talk about reaching what they call artificial general intelligence, where the things would be as smart or smarter than humans. And so then does it revolutionize industries? Does it create jobs, booms and industries? We don't even know. That's what the AI companies would argue, but it's all very far away.

Here's the big dilemma, and I'm not sure you can answer this question, but how does Australia attract potentially these hundreds of billions of dollars of AI investment while at the same time telling companies like OpenAI that it wants tighter controls and basically they need to behave a whole lot better? It's difficult. It's a real conundrum. You know, one of the things that came up in the hearing was copyright law. There were some questions about whether, if Australia had tighter IP protections and copyright law, would that lead OpenAI and others to not want to build data centers and train their models in Australia if it were easier to do it in the US or elsewhere? That's one issue. And you can understand that the government would want to, and its citizens would not want to loosen, you say, copyright restrictions for some of these companies.

Another issue that came up was data centers. And someone asked about a data center construction by a data center company in Tasmania that apparently has faced some opposition. And OpenAI said, well, we're not actually aware of that.

And oftentimes we are just using these facilities that these data center companies build. But they did say, you know, you can imagine Australia being an attractive market for data centers because there's a lot of land, there's a lot of space, there are resources, and it is, you know, fairly open to business compared to other governments, let's say, in the region, and there's good infrastructure.

So you can imagine it being a good, attractive destination for some of these companies to build these things. But at the same time, I think they would look at opposition to data centers, they would look at the stance, they would look at what policymakers are saying, and that could deter them, for sure.

Yeah, the circumstances and the geography of Australia does seem like the perfect destination for a data center boom. And you're also right, there is increasing opposition here. There was a data center that just got axed last week in Sydney due to community opposition. It was going to be right near a school and homes.

But look, I know this question is being asked a lot right now in different countries around the world: What could Australia actually gain from a data center boom? You know, is the prize simply a lot of enormous data centers sitting on Australian soil? Or is there a real potential to turn this into something much, much bigger, with higher productivity, you know, loads of jobs and a genuinely competitive local AI industry?

I think there is that potential, and I think it's really easy to understand why people are nervous or skeptical of some of these constructions because the technology, the gains really do seem far off.

Now, in our day-to-day lives, if you're working in an office or if you have, you know, lots of friends and family who are working in the knowledge industry, you're hearing the best use cases right now for this technology. It's people using them to make their office tasks easier or to save time in creating PowerPoints or helping create documents faster or do deeper research that you couldn't do.

And I think that people who are maybe using it in that capacity would be the most prone to seeing how it could be useful over the long term.

And even among many people who are using it now in the office, there's skepticism about how fast it will develop and whether it'll really benefit them and not the larger AI companies.

It's a very difficult time, I think, for big US tech firms. In the wake of some of the revelations about social media's harms, particularly to children, it's a really difficult time for big, particularly American tech companies to try to convince the world that they are building world-changing products that are gonna benefit humanity.

That's what they argue. It's a really difficult time for them to say that, particularly when there's this visual manifestation of these huge data centers being built that appear to require a lot of resources and whose benefits seem far away.

It's a very difficult time and I think that Australia, like many places, is thinking about: What will this technology actually do? How will it benefit? What are the costs to us and to the environment? And I think that a lot of the potential gains seem quite distant.