CROZIER, Mr Chris, Chief Artificial Intelligence and Information Officer, Defence Digital Group, Department of Defence KEARNAN, Dr Sheridan, First Assistant Secretary, Strategic Policy, Strategy, Policy and International Group, Department of Defence TAILBY, Mr Duncan, Acting Chief, Air and Maritime Division, Defence Science and Technology, Department of Defence [14:07] CHAIR: Welcome. I understand that information on parliamentary privilege and the protection of witnesses giving evidence to Senate committees has been provided to you. You have the opportunity to provide some opening remarks, and at the conclusion of those remarks we'll get under way with questions from committee members. Mr Crozier : We don't have any opening remarks. CHAIR: Can you provide a general observation or overview of the intersection between national security and artificial intelligence? Dr Kearnan : Artificial intelligence will touch all parts of government. There will be a particular framing for the national security parts of the system that might have unique interests that need to be managed and protected. Artificial intelligence is a broad range for defence; it touches the whole enterprise and the side of the house holding how the department actually runs, but it also has a unique aspect in the function of defence and how artificial intelligence will work in the lethality of defence. Mr Crozier : There are two dimensions: there's an offensive and a defensive component that defence needs to be aware of. On the offensive side, I'm talking about our adversaries utilising it, particularly with things like cybersecurity and so on, which means we have to be utilising it on the defensive side to make sure that we have equal or better technologies and solutions. Likewise, there's the offensive side, ultimately ending up in kinetic effect. CHAIR: Defence released a 'responsibilities of AI' policy last year. Can you provide some details on what that covers? Dr Kearnan : The policy covers how artificial intelligence will need to be managed within Defence. That's across the enterprise as well as for the lethality. It talks through the values based principles that Defence needs to consider for the introduction of artificial intelligence. I can step through those. There are basically five key ones. It needs to be managed in terms of accountability. It needs to be able to deal with bias and harm mitigation. It needs to be explainable. It has to be human centric and also reliable and secure. It details some of the control measures that will be factored in. That ranges right down from legal and policy through to actually getting to the use of artificial intelligence, how you design and develop it, test and evaluation, predeployment et cetera. CHAIR: Are there any considerations with respect to AI regulation and legislation that the committee should be aware of that are unique to the national security landscape? Dr Kearnan : It terms of regulation, I don't think there are considerations of legislative or regulatory functions at the moment. A lot of those considerations will be worked through the whole of government. The Office of AI in PM&C is obviously dealing with the AI standard. I think that's going to set a lot of the frameworks. The whole-of-government-system is very connected on the progression of AI. Whether in the future it is a requirement will be worked through. CHAIR: In terms of a sense of what is unique to national security, what you're saying is that that is being worked through as a whole-of-government process. Dr Kearnan : Yes. CHAIR: Deputy Chair. Senator DUNIAM: In terms of that process you talked about before, the five steps, who makes the final decision on whether AI can be part of any process within Defence? Dr Kearnan : It basically is managed right through the entire life cycle, whether it's enterprise systems or weapons systems. What has had to be developed is a system where anyone who is working—because multiple people might work on the introduction of AI into Defence through that life cycle. There could be an earlier phase, which DST might be involved in—that's more the test and evaluation—right through to actually getting it to a point where operators are using it. Senator DUNIAM: Is that a gated thing? If it gets passed this early-phase element, then it's into another stage. It might die at gate 1 or not be approved and move onto the second phase. Mr Crozier : Part of the new role, the chief artificial intelligence officer role, is being able to actually provide that portfolio view of the use of AI or the potential use of AI starting from gate 0, as you've highlighted, all the way through to delivery. This is a new formation, a new element. We will work with our vice chief of Defence and his team in terms of projects and programs. I think it can be anticipated that all technologies moving forward will have an AI component. Whether we determine that we're going to turn it on or what portion of that we use will be dependent on Defence's policies, procedures and requirements. Senator DUNIAM: At any point in time related to the question? Mr Crozier : At any point in time throughout its life cycle. Absolutely. Senator DUNIAM: So the decision you mentioned—was is it the chief of Defence? Mr Crozier : The vice chief of Defence. Senator DUNIAM: Is he or she the ultimate decision-maker, or is it someone else? Mr Crozier : The authority in relation to the application of the technology will go through a series of assessments. At this point in time, there are two critical assessors: myself and our chief cyber and security officer, our CYSO. What we're finding with AI technology is a critical element around the artificial intelligence. We need sovereignty. We need to ensure that it's not calling home, if it's an offshore. And many of these technologies are offshore. We don't want leakage and so on. So the two critical factors are, firstly, the cybersecurity officer making sure. Then ultimately, at this point, my role is to make sure that we're compliant. Senator DUNIAM: Mr Crozier, when was your position or that officer's— Mr Crozier : Seven weeks ago, fresh off the ground. Senator DUNIAM: Ground running! Mr Crozier : Yes indeed—absolutely. Senator DUNIAM: Excellent. You raise an interesting point there, and I was going to come to it a bit later on. Obviously, the topic we're discussing here more broadly is the pathway to have the frontier labs and that capability in Australia and what the roadway to that looks like. You talk about some cybersecurity issues related to them not being based here. Do you have a view on the benefits that might be available to us as a country, from Defence's point of view, of finding a way to engineer a partnership so that we do have that capacity and that power based in Australia? Mr Crozier : Yes. It's a great question. We've been very cautious. We do not want to be using AI toolsets that are located offshore. The companies that are involved in the generation of that—absolutely there are, obviously, tier 1 organisations. But we want the compute, we want the application, we want the data and we want the decision-making all based in Australia. So we've been very conservative in the adoption of the technologies because we have found that, oftentimes, that's not the case for the Department of Defence. The companies are offshore and they have not brought it onshore. So we've made it very clear. If you want to be in the game with us, you have to meet those criteria. Senator DUNIAM: What are the risks? You've got to send stuff through a cable under the Pacific Ocean back to the United States to take advantage of their capacity. That, I assume, is one of those risks you're talking about. Mr Crozier : There are two immediate risks. The first is leakage of national information. We do not want anyone offshore being aware of what we're doing, whether it's partners or adversaries. That's just not acceptable. The second one is, obviously—and people will talk about it—the kill switch. We simply can't be working with technology that's based offshore where someone can have a bad day and turn it off. We need, particularly when we're getting into a conflict, to be 100 per cent capable with our military. Senator DUNIAM: And we can only do that if it's here. Mr Crozier : Correct. Senator DUNIAM: Do you have any views—as a last question; I know we're tight on time—around reporting of incidents and maintaining some sort of visibility or control over, as we learn, when things don't go according to plan with AI? Do you have a view on that, given the sensitive nature of what you do? Mr Crozier : I think that's a really, really good point. There are obviously sensitivities. And there are degrees of sharing, particularly amongst Five Eyes, that would obviously not be made in public. Take, for example, the Hugging Face instance we saw recently. That was (a) shocking but (b) excellent in that it was actually then provided, in terms of transparency. It was not hidden; it was made available. All the logs and so on are there for people to be able to research and so on. We can anticipate that amongst the Five Eyes there's an opportunity to share learnings around the technology sets. We discuss interoperability across the five, which means, oftentimes, that you're using technology stacks which are very similar. The ability to be able to learn from one another is an essential part of being in the five—but not doing it in a manner where it's then made publicly available. Senator DUNIAM: Understood. CHAIR: Senator Pocock. Senator DAVID POCOCK: Thank you for your time; it's very good to see you. Data centres, especially the large ones that we're talking about—we talked earlier about the two-plus gigawatt one going into Queensland—could clearly be the target of any future conflict. Is Defence mapping this out? What contingencies are being put in place? Mr Crozier : It's a really excellent point. If you look at the recent examples in the Middle East, the data centres were targeted. You are aware, Senator, that we recently signed a contract with Google for a Google disconnect to a hyperscaler cloud. That is secret, and that is a meshed series of data centres. I can't go into how many or where they're located, but there is a substantial degree of redundancy specific to this. So we have designed from the outset—we hope we never get there, but, in the event of kinetic, we want to (a) heighten the noise and (b) spread across multiple locations for survivability. The Ukraine conflict has highlighted the necessity for that, and the Middle Eastern conflict has highlighted that as well. Senator DAVID POCOCK: We hear a lot of talk about the need for sovereign capability. Our secret cloud is provided by Google. Is that a concern? Why aren't we seeing more investment in Australian sovereign capability? Mr Crozier : To a degree of sovereignty. First of all, I will make the point that the Google disconnected cloud is here in Australia. It is hands on by Australians. It is managed by Australians. It was built by Australians. Yes, the underlying compute hardware and so on is coming from the United States. We are leveraging the best new world in terms of the Google application stack, but it's actually Australians that are managing it. The second point is that it's disconnected. Somebody can't wake up one day and, having a bad day, decide they are going to turn it off. It is sitting here in Australia, not connected back to the United States. Thirdly, as I highlighted earlier about our mesh network, considerations were made in the actual execution of contract to ensure that what you're highlighting simply couldn't occur. Do we need sovereignty sitting on top of that? In the case of AI, for example, we need AI engines that are, again, controlled and managed by Australians as part of the stack. The Google ecosystem allows for exactly that. We get a fantastic fabric. We get it on a mesh basis. We get it on a survivable basis. But we are not just using Google. We are using a variety of other companies and other technologies to provide military capability. Senator DAVID POCOCK: But we are still relying on foreign companies to provide this sort of capability. What's being done to actually ensure that we have sovereign capability? I know Defence has a different definition of 'sovereign capability'. My understanding is that it's basically anyone who can sell to another military. That is different to the Department of Finance's. But what work is being done to advocate for a truly Australian owned, retained economic benefit, 'pay your taxes here in Australia' kind of understanding of sovereign capability? Mr Crozier : As I said, we are working with a portfolio of companies—Australian and non-Australian—on a variety of different technologies, artificial as well as non-artificial intelligence based technologies. What I would highlight is our ICT technologies in Australia will still rely heavily on predominantly US based firms like Dell, HP and Google. So the underlying element is always going to be non-Australian. But the fact that it's in Australia in data centres that are managed by Australians on a mesh basis, it's disconnected and it allows Australian companies to put their applications on top of that is what we're seeking and making available. Senator DAVID POCOCK: Why are we seeing European and Nordic countries becoming a lot more wary of reliance on US tech giants, and are we seeing any concerns raised in Australia by the government or the department? Mr Crozier : I am probably not best placed to answer that, other than to say that we have a tremendously positive relationship with US firms and with the US itself. So we are leveraging what we see as the benefits associated with that. Senator DAVID POCOCK: How much is Defence investing in developing sovereign AI capability? Mr Crozier : Going forward—and this is, again, part of my role—we are reviewing the IIP investment portfolio. There is well north of $1 billion that is being identified as a potential requirement for AI in the forwards. So we will work to be able to leverage and utilise that within the most appropriate portfolio technologies. Some of that will be Australian based, but some of that won't be Australian based. So it will be there or thereabouts. Senator DAVID POCOCK: Is there a target for Australian based— Mr Crozier : We are working through that. As I said, I'm seven weeks into my role. We will be working through that in due course. Senator DAVID POCOCK: What work is the department doing to advocate for internationally coordinated approaches to AI safety for the highest-risk AI? Dr Kearnan : The department works with the Department of Foreign Affairs and Trade on a range of international areas. There is the line of effort for the responsible use of AI in the military domain. We have been leaning heavily into engaging in that space. Senator DAVID POCOCK: In terms of safety, are you looking at some of the ideologies of the owners or CEOs of these tech companies that you are relying on, or is that outside the scope of what Defence looks at? Mr Crozier : I think there is always consideration. At the end of the day, there are always consideration taken. We have had discussions in terms of elements, individuals and so on. They are discussed at times, yes. Senator DAVID POCOCK: Does it guide your procurement process or do you just discuss them? Mr Crozier : There are degrees of caution, shall we say. Senator DAVID POCOCK: I'll chuck a bunch more questions on notice to get a bit more detail. CHAIR: I appreciate that. I understand Ms Mascarenhas has a couple of questions. Ms MASCARENHAS: I note that the National defence strategy says: The most significant potential for technological disruption in the coming years may come from artificial intelligence … I'm keen for you both to expand on that, but the other thing that I'm curious to understand is the tension between government accountability—big bureaucracies—and having the agility to respond to fast-moving technology. How do we balance being agile and able to adapt with making sure that we have appropriate accountability mechanisms in place? I have a follow-up question for after that. Dr Kearnan : The National Defence Strategy definitely identifies artificial intelligence as sitting within autonomy and autonomous systems. Autonomy has been around for quite a while, but it will speed up. You just need to look at recent conflict, like what's happening in Ukraine, to see the amount of autonomous systems that are getting used. The nature of warfare is changing at a fairly fast technological pace. Mr Crozier : I think one of the critical factors is the time for decisions—what we've seen, particularly in Ukraine and the Middle East is that the decision speed is now down to seconds as opposed to days or weeks, and that can only be achieved through the appropriate application of artificial intelligence. As I said, there is offensive and defensive. We need to be operating at both levels with artificial intelligence moving forward in the Department of Defence. Mr Tailby : If I may add, the Defence Innovation, Science and Technology Strategy complements the National Defence Strategy. It has AI as one of the six priorities within that strategy and looks at decision advantage—how we can use these technologies and the data across a range of systems and capabilities in Defence to gain that advantage. Ms MASCARENHAS: The additional thing I'm trying to get my head around is whether our internal decision-making frameworks are compatible with the use of AI and innovation internally and whether our decision frameworks will move fast enough. Mr Crozier : It's an excellent question. I think this is, again, part of the reason for my role—to be able to move forward on the basis that artificial intelligence will continue to mature. What we see today will not be what we see in 12 months time. We need to ensure that the way in which we're applying our policies, processes and procedures is keeping up to speed with the capabilities that we have and, more importantly, keeping up to speed with our adversaries. At the end of the day, there are adversaries out there. We need to make sure we have the capability of defending ourselves. Ms MASCARENHAS: The other thing I am curious to understand is the tension between—I'll use this term—cheap technologies used in warfare versus our very expensive kit and the role of AI. How do we get the balance of cheaper technologies used in warfare versus very expensive technologies right—and the way that AI can bridge the gap in relation to do those? Do you think that we've got the ratios correct for that? Mr Crozier : That's a really excellent point. It's something that the Ukraine and Middle East conflicts have highlighted—large, exquisite solutions are oftentimes thwarted by swarms of very cheap technology that's underpinned by AI. We are working through that. Obviously, we have ASCA, our accelerator for technology, we have DSTG, and we have capable Australian companies. So we are able to learn, from a distance, lessons out of those conflicts, to the benefit of this country. Mr Tailby : We also undertake a lot of analysis around what the mix of those different capabilities might be, using modelling and simulation and undertaking experimentation on their effectiveness in a range of different operating environments. CHAIR: We'll bring that to a close. Thank you very much for making yourselves available to come and provide some evidence for this committee's work. I understand there might have been a couple of questions on notice. If you are able to provide them, the committee has agreed to responses to those by 26 September 2026..