Radio Interview, ABC Radio National

SUBJECT/S: Artificial intelligence

HOST, SALLY SARA: Richard Marles is the Acting Prime Minister and Minister for Defence and joins me now. Minister, welcome back to breakfast.

ACTING PRIME MINISTER, RICHARD MARLES: Morning, Sally. How are you?

SARA: Very well, thank you. Was the government aware of this breach prior to notification from OpenAI?

MARLES: No. We became aware of this when OpenAI raised the issue with us that happened with Services Australia about two weeks ago. Ministers were informed at the end of last week and over the course of the weekend, and we've just taken the last few days to assure ourselves as best as we can, as the Prime Minister's just indicated that the impact, if you like, of this incident is relatively minor, in as much as no personal information has been accessed here, the system itself hasn't been compromised. That said, this is a really serious incident. I mean, what we've got here is an artificial intelligence agent which has, in an unintended way, gained unauthorised access into an Australian government website. It's obviously the first time that's occurred in respect of the Australian government's IT systems, and that is why we're taking this seriously, continuing to investigate exactly what has occurred here, and we are working cooperatively with OpenAI in relation to that. But we regard this as a very serious incident. And as the Prime Minister has indicated, we've established a task force being led by the Department of Prime Minister and Cabinet to investigate this as thoroughly as possible.

SARA: Has OpenAI given any information about the context here, why an OpenAI agent performed this task? Did it escape a sandbox? Or what was the task that it was given that it ended up doing this?

MARLES: Yeah, look, they have. So, this is a- they're developing obviously more sophisticated artificial intelligence models. In the course of that, an AI model was given a benign task of researching health and medical statistics. It approached four Australian government websites- or Australian websites- I should say two of them; one is the Victorian Department of Health, another was the NSW Government website, and then there was the Australian Institute of Health and Welfare. In relation to those three, it interacted in a way that a member of the public might, so it only acted in an authorised way. But in relation to the medical portal of Services Australia, it sought information, information was not given, and then it effectively hacked in to that medical portal and got that information anyway. And that is that unauthorised access, which we are very concerned about. The information is benign, in a sense, and as I say, the impact is relatively minor, but the incident is very serious. I mean, as we've seen, an artificial intelligence agent, in an unintended way, access an Australian government website in a way which is unauthorised, and that is obviously deeply unacceptable.

SARA: What more do you know about the size, the volume of the data that's been compromised here?

MARLES: Look, we're still working through all of that. I don't think we're talking about a lot of data in the sense that it was essentially asking a question and seeking specific answers. But to be clear, we're still investigating all of that and we are working with OpenAI in relation to it, and they are being very cooperative in respect of that. I mean, we need to understand everything we can here. But we felt over the last few days, we've wanted to assure ourselves that, as best we can, that the impact of this is relatively minor and make that part of what we are informing the Australian people about now. But we regard this event as very serious and we're taking it as such. And we've expressed our deep concern to OpenAI, as the Prime Minister indicated in the clip that you played.

SARA: You're listening to Radio National Breakfast. You're hearing from the Acting Prime Minister and Minister for Defence, Richard Marles. Is there a penalty here, broken the law with this breach?

MARLES: Yeah, that's a good question. And again, that's one that we are working through here. And so we will look at what is the legal situation in respect of this and what it means to have gained an unauthorised access, albeit in an unintended way. So, that is one of the questions that we're continuing to investigate through the task force.

SARA: Does this raise questions or concerns about the capabilities of Australia's security agencies that we had to wait for OpenAI to pass on the information we didn't know that a tripwire had been triggered?

MARLES: Well, look, again, all of that is being investigated by the task force that's been established, to be clear. But what I really want to emphasise in relation to that question is that what we're talking about is information which is at the lower end of sensitivity. I mean, we're not talking about the very highly sensitive information that we have in respect of national security. And obviously, in relation to that, we have, you know, enormous security measures which are in place around them. So, this is a portal which would have had relatively low levels of security, which befitted the fact that the Information that it contained was not particularly sensitive. So, people should be clear about that. In a sense, you know, we keep our most important national security information behind a fortress. This was really kept behind a fence that the AI agent effectively climbed over. And so- but we do need to be really clear as to what's happened here. We continue to work through that and to make sure that all our systems are fit for purpose. I am very confident, as the Minister responsible for the Australian Signals Directorate, that our key national security information is very tightly held and is very secure. And, you know, the Australian Signals Directorate, which manages all of that, is really one of the very best agencies of its kind in the world. But there's a lot for us to work through here and we certainly need to learn all the lessons we can at our end. But I think important today, what we really need to be seeing in all of this, I mean, this is a warning about the fact that artificial intelligence, which is a technology that has huge opportunity to benefit humanity, has to be developed with enormous care. There have to be guardrails and safety measures in place which are way ahead of the capability which is being developed. That's what the Prime Minister has called for along with other world leaders at the United Nations this week. To be fair, it's actually what the tech company CEOs have been saying themselves to the UN overnight. That's what Dario Amodei the CEO of Anthropic, said in his essay a couple of weeks ago. We do need to have guardrails in place. We do need to be developing this technology with extreme care. And I think that's the point that we should take from this incident.

SARA: Richard Marles, thank you for joining me this morning.

MARLES: Thanks, Sally.