HAWKINS, Ms Zoe, Co-Founder and Deputy Executive Director, Tech Policy Design Institute WEAVER, Ms Johanna, Co-Founder and Executive Director, Tech Policy Design Institute [12:10] CHAIR: Welcome. I understand that information on parliamentary privilege and the protection of witnesses giving evidence to this inquiry has been provided to you. I invite you to make some opening remarks before we get underway with questions from committee members. Ms Hawkins : Good afternoon, Chair, Deputy Chair and members of the committee. Johanna and I welcome the opportunity to provide evidence to you today. As a non-partisan, independent think tank, the Tech Policy Design Institute exists to shape technology for the benefit of humanity. One of the ways that we do that is by developing independent research that can serve as an evidence base for important policy processes like this one. I will focus our very brief opening remarks on the dual concept of AI sovereignty and AI agency. I know there's been a lot of discussion about AI sovereignty before this committee and rightly so. All nations are sovereign, but their power to exert that sovereignty on the world stage surely differs. Consider, for example, the power of a Pacific island nation and the power of the US. Both are sovereign but clearly wield different levels of power and influence on the world stage. At TPDI, we've developed a methodology to define and measure AI agency, which is the power of a country to exert its sovereignty on the world stage and to shape its future, particularly as it relates to AI. You can think of AI agency as the way that Australia can adopt AI whilst making sure that it aligns with our national interests and doing it on our terms. That's the concept behind AI agency. Sovereignty is often framed as a binary: you either have it or you don't; you're either self-reliant or you're dependent. When it comes to AI, that's not a very helpful way to frame the policy problem, the global complex supply chains that sit behind artificial intelligence and the large number of distinct AI capabilities that make up a thriving AI ecosystem. Our AI agency tool expands the concept of sovereignty from a binary to a sovereignty spectrum. This spectrum recognises, firstly, that international access to world-leading technology does give a country power as a starting point. Secondly, we must have sovereign control over key capabilities. That's a must. Thirdly, having resilience actually comes from reliable and diversified choice between different sources of that capability, both international and sovereign. Fourthly, and in a best-case scenario, a country actually has such sophisticated levels and surplus of a particular capability that it has export leverage, so other countries actually depend on it for that capability, giving it strategic advantage. When you zoom out and apply this frame of access, control, choice and leverage across the full list of 103 AI capabilities that we've defined in this AI agency tool, it allows you more levers to pull. It gives Australia and the committee more options and visibility from a policy design perspective. It gives you the full chessboard so you can see the next strategic move to make based on our unique circumstances as a country. We have used this methodology, which was developed in consultation with more than 250 experts around Australia, and applied it to see where Australia is today, mapping Australia's maturity, sovereignty and agency against all 103 AI capabilities. This structured assessment can be used as a basis for measuring and monitoring Australia's sovereign AI capabilities going forward. We hope that this framing of AI sovereignty as a spectrum and AI agency as the policy goal is useful to the committee's considerations. We'd be happy to answer questions about the assessment and any other issues of interest to the committee. CHAIR: Thank you very much. I might just kick off flowing on from that. It's a really interesting concept about AI agency and the spectrum of sovereign capability. We have heard from previous witnesses, and I think this is consistently raised in the national debate, around the opportunity that Australia has with regard to the frontier model hyperscalers that are keen to come here, keen to invest and to bring a proportion of their compute for training their frontier models to Australian shores. Can you provide your assessment of that connection or of taking that opportunity. What does that mean? To what extent does that flow into the work that you are doing at the moment around the assessment on sovereign capability and agency? I like the idea of the 103 AI capabilities. I also suspect that that's not a ceiling that we've reached. I suspect there'll be more as this technology continuously evolves. I'd really be keen to get your views on what's in front of Australia right now, that concept of securing these frontier models for training here and how that relates to our capacity for agency and sovereign capability. Ms Weaver : Obviously, there's a lot of focus at the moment on frontier capabilities and Australia securing those capabilities onshore. That is one of the 103 capabilities the assessment looks at. We made the determination that it doesn't make sense for Australia to develop our own frontier models. The investment that that would require is enormous. Just to put it in context—it is hard to get reliable figures—roughly speaking, the amount that Anthropic spends on compute for nine months of training is roughly the amount that the Australian federal government spends on public order and safety. Of course we could do it, but it comes at a significant opportunity cost. The idea of securing access—if we think about the elements of agency, sovereign control being one and international access being another—does give Australia a significant advantage to have these capabilities onshore, particularly from a national security perspective, and access to those models early, but that's not something that will happen unless we negotiate for that access. I think that's something that I have seen in the committee's letter that has been recommended. We definitely endorse that approach; that if they are on our shore, our intelligence community should have early access so that we can be having, for example, patching of security vulnerabilities and other aspects of that. I think, though, when we look at the frontier capabilities, we also need to look at not just the large language models but also the other national security benefits that come from that. There are, of course, significant national security benefits that others have spoken about extensively to the committee, but a lot of Australia's other defence and national security capabilities will also depend on advanced AI capabilities. The analogy that I like to think about with this, particularly with the frontier labs, is a port analogy. In the 19th and the 20th centuries, securing a port meant that you not only had the foreign ships and the Australian ships passing through that port but also the economics, the whole businesses and industries that pop up around those ports. I think there's every indication that these frontier models will be the equivalent of a port of the future. There's a real national security advantage of having those models, but there are also economic and broader benefits for Australia to be secured from that as well. CHAIR: I think that captures quite a lot. I hadn't thought about that concept of a port. We've been getting some really good analogies. It's been fantastic. In terms of that, and in terms of the work that you're doing in your space, where do we need to be getting the settings right in order to, as you say, build up that capability and agency, but do so in a way that continues to bring a sense of trust and confidence from the Australian people in this technology and all of the infrastructure that sits behind it? Ms Weaver : The research really indicates that it's obviously not going to be sufficient for Australia just to have access to frontier models. We need to develop our own industry and our own capabilities. For us, the assessment is prioritising an investment in public interest compute. I know there's been quite a lot of conversation about compute reserve, and I'm happy to speak about, perhaps, some other alternatives or complements to that if that's of interest to the committee. I also think we really need to be leaning into—sorry, when I say 'I think', I mean the assessment reveals that the areas we should be leaning into are also skills. This is expertise for AI-specific skills but also general skills and skills to ensure that we have inclusive and discerning adoption of AI across the ecosystem. Also, separate bodies of research, which Kate Chaney, the member for Curtin, was referring to earlier, around trust really indicate that Australians—there's a lot of conversation at the moment around Australians not trusting AI. We looked at 'well, why don't they trust AI, and what would it take for Australians to trust AI?' The resounding response—this was a national, representative survey—was that they want to feel that it is well governed. Seventy-five per cent of Australians have said that they would adopt AI if they felt those guardrails were in place. So, I think, regulation is so often presented as red tape or a barrier to adoption, but, actually, to the contrary, what the research is showing is that, if we can demonstrate that these guardrails are in place, we will encourage and enable adoption. They're just some of the areas that I think we should be leaning into. Ms Hawkins : I'll just add to what Johanna mentioned about that enabling nature of regulation as a tool for building Australians' trust and that willingness to adopt. It's just recognising that that is another capability that's listed in the 103—that concept of whether we have the regulatory frameworks—and whether we have the capability to enforce them as well is a separate capability. So it's recognising that, actually, to deliver on that sentiment requires all these different components, and it's an area that the assessment revealed warranted additional attention—looking at what Australia's regulatory settings are and what our capacity is to enforce. Clearly, that's been front of mind in the public debate over the last year and brought to the forefront over recent developments as well. So I think it's breaking down that typical narrative of the idea of 'innovate or regulate', and I think what that national survey and research really found is that that's a sort of false binary and that one can enable the other and is an important foundation ingredient to the whole AI stack. CHAIR: Yes. I appreciate that. Deputy Chair. Senator DUNIAM: The 103 capabilities you've got in your assessment tool—how did you come up with that list? Ms Weaver : It's a very good question. We actually started this body of work because Zoe had done some work with partners at Oxford, looking at mapping global compute. It was around the conversation that we were having before the National AI Plan. I said, kind of blithely, 'Oh, surely we could articulate the elements of national AI capability,' and Zoe kind of blanched. Senator DUNIAM: No pressure! Ms Hawkins : I fell off my chair, actually! Ms Weaver : Yes, exactly! We embarked on a specific research project to do that. We held consultations around the country—13 different roundtables—and consulted, as Zoe said, with over 250 different experts. It covers everything from infrastructure through to data, models and applications and then also looking at skills, governance, and diffusion and innovation. So it really covers the full stack. That was the result of those consultations, which also helped us to develop the methodology, the assessment method and the maturity rankings. Ms Hawkins : Just to reinforce that, the findings of this assessment are not purely our own. Certainly, both the framework development, that list—it was 101 and then became 103, which I was disappointed about because 101 sounded so much better! But we followed the experts, and they said, 'There are two missing.' It was not only to define that structure, which I think, as I said, is a repeatable method that other countries have indicated interest in as well, because every government and country is struggling with this question of 'what are the ingredients that we're looking for?' The experts also then helped us do that assessment of where Australia stands today against that 103. So it's certainly not just TPDi's assessment; it is shared by the multistakeholder expert group. Ms Weaver : For show and tell— Ms Hawkins : It's always good to do! Ms Weaver : this is an example of one page of that assessment. CHAIR: Wow. Ms Weaver : What is particularly relevant, I think, for the committee is that it's not only defining those 103 capabilities. We're also giving examples of those capabilities and indicators of those capabilities. It's really breaking it down into making concepts like a data centre—we're often talking about a data centre. This breaks a data centre down into 27 component parts. That's not because we're trying to make it difficult; it's because each one of those component parts is owned usually by different organisations. If we're really looking at a question of sovereignty or agency, which of that do we need access to, which of it do we need to own, which of it are we good enough to export and where is our choice? You need to have that level of nuance, but we also wanted to make it accessible. This is something that maybe a nerdy and interested person off the street could pick up and, we hope, actually understand, whether it's via the example column, the indicators or the definitions. Senator DUNIAM: Is that work you've done available online? Ms Weaver : Yes. Senator DUNIAM: I'm very interested in this. There have been these nebulous answers provided to most questions asked around sovereignty. I know it's a spectrum. It's not a hard and fast 'this is sovereign, this is not'. It's all of those nuances you talk about, which is very helpful. Ms Hawkins : That's wonderful to hear because that is literally the policy problem that we spotted. Why we did this work was both because AI was being spoken about like it one thing despite it being all these different capabilities that come together and because that binary of sovereign/not sovereign doesn't leave you with much nuance for real policy solutions. This is intended to do that. Beyond the beautiful PDF which Johanna just held up, we also really wanted to make it usable for policymakers, so we've digitised it into an interactive dashboard. Senator DUNIAM: Is it on your webpage? Ms Hawkins : Yes, ai.techpolicy.au. Senator DUNIAM: We'll find it there. Ms Hawkins : Look for that interactive dashboard. Senator DUNIAM: On where the rubber hits the road, in terms of a government trying to catalyse a sovereign industry, are there any outputs or recommendations that might arise out of the work you do or from the model itself that lead to how we might structure a procurement process to, as I say, catalyse an Australian industry? Ms Hawkins : The assessment shows us where Australia is already leading and the opportunities where we can double down on areas of competitive advantage. Out of the 103, Australia is world leading in eight of those capabilities. That includes critical minerals—which might be of no surprise to the committee—but also on our datasets. Australia has some of the richest datasets in the world, everything from medical datasets, environment and resources, geospatial, demographic and infrastructure data. Five of our eight world-leading capabilities actually sit in the data layer, which is really interesting, as well as in expertise that Australia has in computer vision—so the ability of machines to interpret and visualise the world around them. Lastly, Australia is a world leader around international engagement and global norm shaping. That gives you a starting point. Then what you can do with that is say, 'If we zoom out, what are the complementary combinations of those leading capabilities that could guide where you'll get the highest ROI for government investment—where we're already strong.' As an example, think about the fact that if we have really rich medical datasets, and we're really good at computer vision, that positions Australia well to be a world leader in, say, AI medical imaging. That's the value, again, of seeing the full picture rather than looking specifically at those capabilities. We can speak more about that. Ms Weaver : That's a starting point. Senator DUNIAM: On the assessment model, as far as you know, no-one else has replicated this or done equivalent work elsewhere in the world? Ms Weaver : To our knowledge, this is a world-first methodology to assess the nation's agency but also its sovereign capability. We've had a lot of interest internationally in the model. It is building off the work that Zoe and her colleagues did out of Oxford. Senator DUNIAM: I'm very excited about this. This is absolutely— Ms Weaver : So are we! Senator DUNIAM: Yes, I can tell. Thank you very much for that. CHAIR: Ms Chaney. Ms CHANEY: I want to start by thanking you for the amazing work you do. I think the AI Agency Toolkit is a useful and nuanced piece of work. I commend you for that. No doubt you read our letter to the PM this week. I'm interested in your views on whether there are any safeguards that are missing from that letter that the committee should be covering. Ms Weaver : I think we definitely endorse all of the safeguards that were included in the letter, particularly the additional funding for the safety institute—I think it's important for us to have on the record that we support that—the independent evaluation, the safe harbour for incident reporting and also the legacy systems review. I think some of the things that were missing for us are some of the issues that have been raised today around privacy reform. I actually think this is the single most important piece of legislation that can pass and have a huge impact on AI regulation. The digital duty of care, which of course was introduced to parliament this week, is also very, very important. And I also think the automated decision-making in government legislation, which we are waiting for—I know, Kate Chaney. you've introduced a private member's bill on that. I think it's really important that we're also looking at government use of artificial intelligence; that is really important. One other area, which wasn't included, that I think is a very important safeguard for Australian democracy is looking at the use of AI in elections. I appreciate that this is a challenging space. We're effectively asking you to regulate yourselves. But I do think it is important that we have a commitment from political parties and from all candidates to have transparency in the use of artificial intelligence and. in particular, the labelling of any products that you're using that are intentionally deceptive. Effectively here I'm talking about deepfakes. Some people call for the banning of deepfakes during elections. I think that would be an ideal position. If we don't move to banning, then there should be, at least, a commitment, if you are doing a deepfake as a political party, to label that. I think that is a bare minimum that we should be having in place. I also think we should be using the forthcoming election campaign—which I don't want to think about yet because we've still got plenty of time to do really good work in this term of parliament—as an opportunity to have a widespread campaign teaching Australians to spot authentic content. We've moved past a point of being able to teach people to identify fakes. It's too hard; the AI is too good. Let's teach them where to go and how to identify reliable, verifiable information, whether that be through a marking from a political party or verification through trusted public interest media. There are plenty of ways we can be teaching people to spot the authentic content, rather than spotting the fakes. So I think they are some of the safeguards that we're missing. The final thing I'll say on that—well, actually, I have maybe two points. One is if we are relying on existing laws—whether that's privacy or whether that's workplace health and safety—to regulate artificial intelligence, we need to make sure that we're funding and resourcing the regulators to be able to do that. I think we've got a lot of focus on the fact that we need to use those existing laws. We haven't had much focus on funding the regulators to be able to do that. I also think we should be specifically looking at the question of liability, as well, around AI agents in particular. Some of the aspects of liability can be managed through existing provisions—for example, saying to a company director: 'Set the expectation that, if your company is using an AI agent, you have an obligation to make sure that agent is doing the right thing and obeying Australian law.' This is quite similar, I think, to the provisions and the expectations that were set around cybersecurity of companies. Directors have an obligation to ensure cybersecurity is in place. I think we need to make sure that we have that expectation when it comes to AI agents. I think there's also a question about liability for companies that are developing these AI agents, as we've seen recently through the OpenAI and Medicare incidents. And I think there is, finally, a unique opportunity for Australia, potentially, around AI agent identity, as an area where Australia can move forward. Ms CHANEY: Can you talk a little bit more about agentic identity. Ms Weaver : I think what really focused my mind on this was when we became aware that the Medicare site had been compromised. That was because OpenAI had a notification. If Services Australia had identified that themselves, rather than it having been brought to their attention, it would have been very difficult for them to have understood that it was an AI agent versus a standard compromise, and it would have been very challenging, if not impossible, for them to have been able to identify whose agent had compromised the system. There are a number of very nascent proposals being discussed globally around whether or not it is possible to establish identities for AI agents, and it is interesting, when you think back to the early days of the internet, that there are analogies here to when we first started having the domain system, for example. You would have the spread of different .com domains, for example. When it first started, it was researchers to researchers. They knew each other, so they knew who those domains were being issued to. Then, when it became more commercialised—similar to what we're seeing right now with AI agents—some domain administrators made the decision that they would attach identity requirements. For example, in Australia, if you want to register the .au domain, you actually have to provide some evidence that you have a connection to Australia. Usually, that's by proving that you have a business name register, for example. That means that the .au domain in Australia is one of the most trusted domains globally. We're not a domain where huge amounts of scams or spoofing comes from because we have that identity requirement. It's not privacy invading, because it's not saying that this is Jo Briskey's website; it's just attaching it to a particular company, for example, similar to the way that a number plate registration registers a car, but you're not necessarily sure who's in the car or who's driving the car at any particular time. It provides a mechanism for accountability, and I think that, as AI agents become more pervasive through our economy, we will need to be able to verify who the agents are acting on behalf of. We will want to do that anyway in the transactions that we're having. But I think there's also an important accountability element to that, and Australia has been a leader on this front in many spaces before. So I would recommend that, if Australia does not establish an AI agent mechanism, the committee recommend that Australia actively look at establishing this. Senator DUNIAM: I reckon you could be guaranteed of that. Ms CHANEY: Thank you. That's a whole area we haven't really talked about much. I just want to go back to the deepfake issue in elections. You say: ideally, ban AI use in political ads but maybe a compromise position is labelling. I've met with psychologists who say, basically, we've been trusting our eyes for 10,000 years, or longer than that, and even though you might be told something is AI, it still actually changes how you feel about it. Why is there a positive use case for AI in political advertising that means we can't aim for a ban on the use of AI in political ads? Ms Weaver : I think there are lots of ways that AI is used in political advertising every day. Just a general Canva image involves artificial intelligence. So I don't think you want an all-out ban. But that's not really what you're asking. You're talking about its nefarious uses. I think that's where the labelling does become important. There is a lot of research to support the point that, once people see something, they think that it is real. But I think that is why the education campaign is really important. We are trying to have people understand now that the content that you see online is not reliable unless it comes from a verifiable source. That requires teaching Australians that they need to look for markers of trust and authenticity rather than look to try and spot fakes. It is a total switch in the way that we have been engaging with the public to date, but I do think an election campaign where you have a trusted entity like the Australian Electoral Commission, which has played that neutral role for a long time, provides an opportunity to educate the Australian population of that issue during an election campaign, but then, also, for that message and those lessons to have much longer—hopefully—positive benefits across the economy. Ms Hawkins : To add to that, I think that it's also an opportunity to demonstrate leadership—that government is expecting to see this level of transparency and labelling from industry. I think it's a really great way to match that expectation and say, 'We expect that from industry, and here's how parliamentarians, politicians and government actually reciprocate that expectation of transparency for the public.' I completely agree about that concern of being convinced even if it's labelled. I think it's maybe an example of us marrying the policy solution with the pragmatism of what is going to be politically viable as well. You don't have to look far across the online safety and misinformation debates to know that there are going to be large pockets of what society expects. Artificial intelligence, as Johanna mentioned, comes in many gradients of converting an image—sometimes it's a background remover; it even goes all the way through to political satire. If someone made the argument that this is the new form of political cartoon, some people might object to an all-out ban. So I think, if you're looking for a pragmatic solution that would move the dial but maybe not get caught up in some of those more complex debates around political expression, that might be another reason to consider that as a middle ground. Ms Weaver : Yes, pragmatic optimism. Ms Hawkins : Exactly. Ms CHANEY: Thank you very much. I just have a last quick question. It's easy for Australia to feel like all these decisions about the big picture are being made overseas. What do you see as being Australia's role in shaping global norms and standards? Are we kidding ourselves in thinking that we can have some influence, and is there any evidence to the contrary? Ms Weaver : As Zoe said, one of the eight areas where Australia has the highest competitive advantage globally is our ability to engage and shape global norms on these issues. I know that there is a lot of scepticism among the general Australian public about that, but Australians have been engaging on these issues for a really long time. So—whether it's Aurelie Jacquet or Lyria Bennett Moses, who've been leading discussions on standard AI standards at the OECD or at the international standards organisations, or whether it's people like Briony Daley Whitworth, who led negotiations on the cybercrime treaty, or whether we go as far back as 2013, when we had Deborah Stokes, who was the chair of the group of governmental experts at the UN, where Australia led the negotiations that established that international law applies to state conduct in cyberspace—Australia has been leading conversations around the rules and norms in the online environment, traditionally, in the context of cyber—and you can add an adjective to that. That means that countries are looking to Australia to provide leadership in this space. So, yes, the US and China are leading in terms of technology development in this space, but, if you travel internationally and ask other governments who they look to to be the sensible middle-ground voices, especially on tech issues, it will be Australia. I think Australia was the first country to have a cyber ambassador, which evolved into an Ambassador for Cyber Affairs and Critical Technology. That position has largely morphed to become focused on cybersecurity issues. I want to disclose my interest here. I did work for the first ambassador for cyber affairs, and back in that time we deliberately conceived it as broad—as in 'cyber affairs' in the same way we talk about foreign affairs. I think the subsequent cyber ambassadors have very deliberately, at decisions of government, done an excellent job but have narrowed the scope back to cybersecurity. I think there is a huge opportunity for Australia to be playing a leading role, and countries are looking to Australia to play a leading role in this space, but we need to either resource the Ambassador for Cyber Affairs and Critical Technologies more to be focused on these AI issues or perhaps establish a position of an ambassador for AI affairs. Dare I suggest another position? But the point is that countries are looking for Australian leadership, and there are areas where we do need international cooperation on these issues. Talking about incident reporting, that will be much more effective if we do that in partnership with a coalition of countries. Let's learn from the mistakes that we made during cyber incident reporting. I was in government negotiating some of those agreements. If we can get it in place across multiple countries at the same time to be having the same AI incident reporting, it makes it so much more efficient. Let's get that going from the start. AI liability, for example, will be much more effective if we have a coalition of like-minded countries agreeing on how and when we will hold companies to account. So there's real demand for it, and countries are looking to Australia. I think that's an area which I would recommend that the committee look into more. I just want to emphasise I'm not being critical of the two very excellent cyber ambassadors, Jess Wilson and Brendan Dowling. They have focused very deliberately in that area. But I think there's scope for us to be doing more. Ms CHANEY: I appreciate that. Ms Hawkins : I will just add to that. I also think that it's not only the interoperability and the efficiency of that moment that Johanna's describing. I think it gives middle powers the strength in numbers to not feel deterred by the either implicit or explicit threats of retaliation from the US government for imposing regulations that might be perceived to disadvantage American companies. So there's a strategic reason to do that as well as a great opportunity for leadership. Ms CHANEY: Do you see the AI Safety Institute playing a significant role in that work? If so, what resources do they need to do that? Ms Weaver : It could be something that becomes a pillar of the AI Safety Institute's work, and, if so, they'd need a team to be equipped and resourced to do that. That fits naturally, in the sense that the AI Safety Institute already operates within a network of safety institutes, variously named, internationally. I would encourage us, though, to think about it as something that needs to sit within the Department of Foreign Affairs and Trade, because the AI Safety Institute is focused on technical issues. This is a strategic issue. This is about positioning Australia so that we are leveraging our full suite of strategic capabilities, and that does require diplomatic statecraft to be able to do that—which, of course, could be housed within the safety institute. But I think it's a complementary focus and skill set, which would work very closely with the safety institute. In general, I would look at it through the various different cyber negotiations I was involved in, in my diplomatic career, and I would say that, when you have the technical people at the table, it's a very, very different conversation. It is a maturing of the recognition of the strategic impact when you see it sit within a foreign affairs portfolio. That's why it was so significant that Australia was the first country to appoint a cyber ambassador sitting within DFAT. Ms CHANEY: This is a very quick last one. We do need that international cooperation around things like incident reporting. International cooperation can take a long time. How do you think we see that balance? Do we get on with having domestic requirements now, and then feed them into international discussions, or do we wait until there are international standards before we do anything in Australia? Ms Weaver : We have to do both simultaneously. But I would push back on international negotiations taking a long time. After the Christchurch terrorist incident in New Zealand, within one month 15 countries had come together with all of the US social media companies and agreed the provisions of the Christchurch call—that we will not accept abhorrent and violent content online—and agreed the measures to take it down. If countries are motivated—and I believe they are motivated, and I think Australia should be using the Medicare compromise as the catalyst for Australia to be leading a conversation to say: 'This is not acceptable.' We need countries to come together, and this doesn't have to be something that takes a long time. Some negotiations—particularly ones that sit within the UN, for example—do take a long time, but there are things you can do in minilateral environments that can move very, very quickly. Ms CHANEY: Thank you. CHAIR: Senator Sheldon. Senator SHELDON: Thanks very much. You spoke before about the deepfakes in election times and some of the potential solutions in a policy sense, though I appreciate you spoke quite broadly. Have you had an opportunity to look at the South Korean example—in, I think, two election cycles or thereabouts—as to how that played out, and whether that is an example that we should give a lot of weight to or give some consideration to? Ms Weaver : You are stretching my memory a little bit on this one! Correct me if I'm wrong, but I understand the South Korean is a direct ban on deepfakes. It's really going to Zoe's points before, around the pragmatism. I think that is a good measure, but it hasn't and doesn't, in Korea, prevent other parties using deepfakes. For me, it comes back to this: if we're not banning it, then label it. Label it as satire. Label it as: 'We've used AI to do this,' but also teach people: 'If you want to make a decision about who to vote for in the election, go to the relevant political sites to get that information. Don't rely on information that's being fed to you elsewhere.' So it's that flip, in saying: 'We're teaching people to look for the authentic content, to look for the verifiable content, rather than teaching them to spot fakes.' That would be the key lesson that has come out of the Korean example, for me. I also think this is not something that Australia needs to make up. The 2024 Munich Security Conference brought together a number of countries and a number of companies. There's a declaration. There are protocols in place. We're not reinventing the wheel here. This is something that tech companies have come together for that has been rolled out in other jurisdictions. Ms Hawkins : Within the 103, we actually considered, which is what Johanna was just mentioning, making sure that citizens are equipped with that ability to live with AI. That was one of the things that came through really strongly in our national consultation. We talk a lot about national capability. I think we often have very institutional visions of what that means. But really it's actually thinking about national capability at that individual level. Societal and democratic resilience will often come down to how equipped individual Australians are with that critical thinking and with that AI literacy. That's thinking through how this needs to start in schools and public awareness campaigns around elections. There are so many different ways that we need to start building that national capability, but it's at an individual level as well. Senator SHELDON: I'm not sure if you heard our previous two witnesses. They spoke about this. It goes to this question of general skills and probably that point you made, Ms Hawkins, on how we build up capability across the community. They went to two examples of particularly the Occupational Health and Safety Act and the Fair Work Commission. My impression from commentary on this is that, in particular, the Occupational Health and Safety Act has quite a settled sense across the Australian community, business community, workplace and workforce that you abide by those laws. There's an ambition and almost inherent knowledge that, if wires are hanging out of the wall, something's wrong. There's something a lot more elaborate than that. People are aware to have a conversation if something looks like it may be or has a possibility of being dangerous or has a possibility of being safer. 'So how do we not only risk-manage with the risk management I have done but consult with the rest of the workforce or the rest of the company or the rest of the industry on those things that have embarked with government guidelines built by industry et cetera?'—is that the frame? Do you think one of the frames would be an area where we've got a really cutting-edge, already built in hot-wired system of community consultation, engagement and actual voice? Is that an area that has some dual ability for the sorts of things you're talking about? Ms Hawkins : Yes, absolutely. I think that it's really picking up on that duty-of-care concept as well. That's why the relevance of the duty-of-care reform is really putting that expectation on service providers to say, 'In the space you create, you're responsible for the risks that may arise and for anticipating those and preventing them as well.' Ms Weaver : I think also both the Fair Work and WHS provisions are relatively technology-neutral. I think we want to avoid a situation where we're continually updating those pieces of legislation whilst at the same time making it clear that AI is just a part of most workforces now and those WHS obligations extend to the use of AI and the implications of AI—potential harms but also the benefits—and then making sure that we're equipping workforces and regulators. I thought the most important thing that I'd underscore from the comment that was made in the previous session was the example, I think, from Finland where they've provided guidance and information about what that means. I think most people accept that WHS would extend into artificial intelligence and the use of artificial intelligence. What Australians are grappling with is: How do I comply with that? What do I do? So I don't think it's necessarily a situation of new laws. It's about making sure we're equipping the regulators to provide that guidance. Senator SHELDON: Similarly, the Fair Work Commission just recently assisted with the guidelines, but also the commission assisted with the now infamous Commonwealth Bank redundancies replaced by AI. But, to the credit of both parties, after that argument and fight and with the assistance of the Fair Work Commission, they came up with a settlement to retrain and reinvest and also deal with the productivity questions the company was trying to deal with. I understand that they've just come to an agreement on some wider initiatives regarding how to engage with AI with the workforce, going to the education and skills development piece. So, again, that question is about how we adapt existing institutions rather than have something separate. Ms Weaver: I think it's also bringing an element of humility. I hope this isn't insulting. I know that might be hard in politics, but the reality is we're going to make mistakes. We haven't done this before. So it is recognising that maybe there are circumstances where things are done wrong and that cause damage. But how do we then make sure that we are recovering from that and applauding when companies and regulators recognise those mistakes and move forward rather than pointing the finger? I think that's actually going to be a really big part of creating a resilient Australian AI ecosystem. Ms Hawkins: Just to add quickly to that, it's one of the reasons that one of our national capabilities is discerning adoption. Importantly, our framework is not starting from the assumption that more AI everywhere, all the time, for everyone is always the right answer but actually that sophistication and maturity look like giving organisations, workers and different people meaningful critical skills, choice and strategic consideration about when AI is actually going to add value. I think that that will help with that trajectory that Johanna was mentioning. CHAIR: Excellent. That brings us to time. Well finish up there. Thank you so very much for making yourselves available. It's been a really useful contribution to our work as the committee. If there's anything further that you've been asked to provide on notice or there is anything further that you think would be helpful that you would like to contribute, please do so through the secretariat over the next two weeks. Proceedings suspended from 12:56 to 13:40 .