Release details
Release type
Related ministers and contacts
The Hon Richard Marles MP
Deputy Prime Minister
Minister for Defence
Media contact
Release content
24 September 2026
SUBJECT/S: Artificial intelligence
HOST, MATT SHIRVINGTON: Now we want to get more on the breaking news this morning. The Prime Minister an hour ago, revealing Australia has suffered a cyber attack with an AI agent infiltrating a government website relating to Medicare.
HOST, NATALIE BARR: Now, at this stage, no personal information is believed to have been accessed, although digital forensic detectives are still investigating that. For more, we are joined by Richard Marles, Acting Prime Minister. Good morning to you. What information exactly was accessed by this rogue agent?
ACTING PRIME MINISTER, RICHARD MARLES: So, the model was given a benign task. It was, as I understand it, part of the development, the testing of it, and it was to go and do a research task in relation to medical and health statistics. In the process of that, an AI agent part of the model interacted with four Australian websites, the Victorian Department of Health, a NSW statistics website, and then the Australian Institute of Health and Welfare, and a medical statistics portal of Services Australia. It's in respect of the last that unauthorised access was gained. So, in relation to the other three, it's sought information. Only public information was accessed. But in relation to the medical statistics portal of Services Australia, it asks the question, information is not given because it sits behind a wall. And then the agent in effect hacks its way in and gets that information and brings it back. Now, the information, as we understand it, that was accessed is benign- it's not anybody's data. It is literally health and medical statistics. So, the impact of the incident is actually relatively minor. But this is a serious incident, obviously, and one that is completely unacceptable, because what we've here is an artificial intelligence agent, so not a human, which in an unintended way has hacked into an Australian government website. And that, right there is an unacceptable situation. As the Prime Minister has made clear, he's expressed our deep concern to Sam Altman, the CEO of OpenAI, about this incident. And I think if we take a step back here, what this really is, is. Is a warning about the fact that artificial intelligence is a technology which has enormous opportunity for benefiting humanity, but it needs to be developed with extreme care-
SHIRVINGTON: Safeguarded, Richard-
MARLES: Guardrails, the safety measures- yeah, they need to be well ahead of the capability itself.
SHIRVINGTON: Look, the big concern here is that one, the communication of information. So, Sam Altman, the CEO of OpenAI, told the PM about this. It happened in June. He's only told us from two weeks ago when he was notified. We've just completed a census where we've given up so much personal information out there onto the web. This is so concerning for people that have that information out there. Do you know if something like this has happened before? Do we know if we've got more information from OpenAI that it's happened to other government agencies? We're very concerned about what's happening here at the moment.
MARLES: Yeah. So, firstly, we're unaware of this ever happening before, which is obviously why this is a serious and significant incident. I think the point that we'd want to make to reassure people is there are differing levels of security that exist across the government's IT systems. And this is information that was not particularly sensitive. And so the level of security around it wasn't particularly huge, if you like. You know, we've got a fence here which an AI agent has climbed over. When we're talking about people's personal data, people's personal security, that information is obviously sitting behind much greater levels of security. And if you look at, you know, the most sensitive national security information that our nation has, I mean, that sits inside a fortress. So, we are talking about a set, an amount of data, or a kind of data, I should probably say, which was not particularly sensitive, and therefore the security levels around it weren't huge. But that's still-
BARR: This time.
MARLES: You know, a very serious incident, in as much as what we've got is an unintended, I mean, open. I went seeking to do this. This was an unintended consequence of the model that they established. And right there is what we should be concerned about.
BARR: Yeah, okay. Richard Marles.