ANDERSON, Mr Mark, National Security Officer, Australia and New Zealand, Microsoft BLACK, Mr Ryan, Director, Government Affairs, Australia and New Zealand, Microsoft CRAIG, Ms Amanda, General Manager, Responsible AI Public Policy, Microsoft [by video link] FITCH, Ms Liz, Head, Corporate Affairs, Australia and New Zealand, Microsoft GIMBEL, Mr Jordan, Associate General Counsel, Open Innovation, Microsoft [by video link] [15:06] CHAIR: Welcome. I understand that information on parliamentary privilege and the protection of witnesses giving evidence to this inquiry has been made available to you. We'll move on to providing a short opening statement so we can save time for questions. In terms of anything more you'd like to add, please provide a written statement to the secretariat. Ms Fitch : Thank you very much for the opportunity to appear today. At Microsoft our belief is simple: people matter more than AI. Our goal is to ensure Australians can help shape AI, benefit from it and use it to build a more prosperous, secure and resilient future. Research shows that Australians are adopting AI at one of the fastest rates globally, with 41 per cent of the working-age population using AI, but trust has not kept pace, with 44 per cent of Australians reporting little or no trust in the technology. Closing this gap is central to realising AI's economic and social benefits. It requires progress in three areas: security, infrastructure and people. First are security and safety. As AI becomes more capable and autonomous, governance must keep pace, with humans firmly in control—always. While Australia starts from a strong regulatory foundation, we support the committee's work to reinforce existing guardrails and address any gaps through AI standards that are proportionate, consistent and aligned with international approaches. Second is infrastructure. Demand for AI and cloud services is growing faster than the supply of available compute capacity. Meeting that demand is essential to Australia's future prosperity and sovereignty. It must be done responsibly, sustainably and in partnership with local communities. That's why Microsoft signed on to the government's expectations of data centre operators when we pledged an additional $25 billion to expand in-country AI and cloud capacity earlier this year. Third, and most importantly, is people. We know many Australians are concerned about what AI could mean for their jobs and future. Workers need a genuine voice in how AI is introduced, along with the right skills to benefit from it. Microsoft is equipping three million Australians with AI skills, and working with unions to ensure workers are central to this national conversation. Microsoft shares the parliament's ambition that AI works to the benefit of all Australians. Now is the time to ensure our guardrails are strong enough to build trust while enabling innovation. We look forward to your questions. CHAIR: Thanks very much, Ms Fitch. I'll kick off. You mentioned the importance of sovereign capability. I'd like to get a sense of what sovereign AI development capability means to you and its relation to having the ability to host compute and have compute investment here in Australia. Ms Fitch : I might ask Mr Black and then Mr Anderson to answer. Mr Black : On the broader question, Chair, we think that Australia should be ambitious in developing local AI capability. We think that, for the benefits of AI to be broadly spread, they need to be broadly shared across countries, across organisations and across individuals. In that respect, we think that there are benefits in focusing on areas where Australia has competitive advantages but also where it has strategic interests. One of those areas is with respect to compute. Australia has abundant lands and energy resources, and we think that provides great capacity for development of compute that can be deployed for a range of different purposes. That includes AI training, but it also includes inferencing, fine-tuning, research and AI application development. We think about that capacity as being a great foundation for Australia to develop capability here. I might pass to Mr Anderson just to add to my answer with respect to how we think about sovereignty and sovereign capability more broadly. Mr Anderson : I think it's really important that we start to divorce the idea of access to models from access to AI capability more broadly. The model is just one part of an overall AI ecosystem, and, increasingly, we're seeing elements that sit within the middle layer—what we call the harness, for want of a better phrase. Think of that as the area where all of your agents will operate, where your security will sit and where all of your governance aspects will sit. We are starting to see that the model, whilst important—if you lose access to frontier capability, yes, you will reduce your capability. We're seeing that a mixture of models and access to different models actually helps produce the results that you want. A good example of this that I can provide is that, back in March this year, when there were restrictions put on one of the Mythos models—it was restricted outside of the US and US citizens—at that same time we released a capability called MDASH. MDASH is a capability where we will allow customers to take their source code, place it into the model and then look for vulnerabilities and provide back to them a series of vulnerabilities stat ranked by priority and organisations so that customers can take charge. We found, using one of our own models that we developed, a much smaller model called MAI-Cyber-1-Flash, we were able to meet and actually exceed on industry benchmarks that of a single model capability. So I think, from a sovereignty and also a resilience perspective, it's important that we think about AI more broadly because this also gives us the ability to bring in locally created models as well, more specialised models, and how we think about bringing them all together in one ecosystem. There's one other key component of that that's outside of the sovereign capability and the resilience aspect, and that is one of economics—the area of token economics, which is the idea of a token as a monetary unit that you use to pay for AI capability. We recognise that, whilst the larger models, yes, may be more advanced in their capability, you don't necessarily need frontier capability to summarise an email; a smaller model will do exactly the same thing for a fraction of the cost. That's why we think about the whole sovereignty and resilience piece as a broader ecosystem play rather than just relying on single models. CHAIR: Going back to the need for compute or the opportunity for compute—the extent to which Australia is placed to leverage off the fact that we have space, energy and a whole raft of things that are very advantageous for big investment in compute, whether it's training or inference but particularly training. How do you see the connection between that and our ability to then leverage off that to build up our own AI ecosystem and investment in Australian skills and the like? And this goes to your point around continuous access to what we need to have for sovereign capability. Mr Anderson : Sure, I'll pass to Mr Black on that one. Mr Black : As I was saying, I think the compute capacity is very foundational to all of those opportunities. What we're seeing with respect to the build out that we're doing, not just in Australia but around the world—and to Ms Fitch's point in the opening statement—is that compute demand is significant and growing, and our challenge is actually keeping up with that demand. I think Australia's got a great opportunity to have an advantage in that space, both in servicing Australian organisations in the way in which they want to adopt and adapt the technology for their purposes and to support areas like banking, mining, agriculture, health, education—all of these great applications that we do see here in Australia—as well as to support, as you say, AI model training. There is great capacity to do that here, leveraging that compute layer as the foundation. As Mr Anderson pointed out, there are a number of other layers to the tech stack that are equally important and where Australia does have even greater capability, we would argue, to capture economic value, particularly in the applications layer, to develop those specialised use cases for areas where it already has economic strength. I think, yes, compute is absolutely fundamental, and Australia's got the right foundations to take advantage of that. CHAIR: What would your sense be if we missed out on potential big investment in the compute capacity or not have as big a role in that demand that is growing? What are the potential outcomes for Australia if we would miss out on that? What's the opportunity cost is the question. Mr Black : The first I'd say is from a productivity perspective. We come back to that. It's a major challenge for the nation. From the Treasury Intergenerational Report that was put out recently, a lot of that future growth in productivity, which is connected to living standards, has been modelled with a dependence on the extent to which AI delivers those productivity benefits. That fundamentally comes back to the demand question that we were talking about earlier and ability to build that out. That's particularly for inferencing and for cloud. That's what our $25 billion investment is connected to. But then, with respect to AI model training as well—I don't know whether some of the earlier witnesses today mentioned this as well—there are great opportunities to leverage that investment to build your broader innovation capability. When we think about the economic and strategic opportunities of AI, we think about it both with respect to innovation but also adoption. I think the productivity gains in particular will accrue from the adoption. But the innovation, the local economic benefit and the local capability to build here in Australia, which ultimately have more than just economic consequences but also strategic consequences as well, will also depend on being competitive in that innovation layer. I think, at this present point in time, we're seeing a lot of competition in that layer, and there is a great opportunity to move now to take advantage of that. CHAIR: Thanks, Mr Black. Mr Hastie. Mr HASTIE: Thanks for appearing today. Could I turn to priority number 4 in your submission on page 4 which is, 'Strengthen Australia's security and resilience for the AI era.' The first sentence reads: AI capabilities should be considered in the context of strategic competition. Strategic competition between whom? Ms Fitch : Mr Anderson, do you want to take this one? Mr Anderson : From a national security perspective? Mr HASTIE: Yes. Mr Anderson : Most definitely. From the point of view that I think that AI is increasingly playing a key part in national security, for sure. If you think about the increase in how we are seeing AI enabled cyberattacks, to give you a good example on that, last week we released our digital defence report where we saw a significant increase in the number of attacks that were using AI as a result of that. In the majority of cases, that was AI that was supplementing the human capability, so it was doing the things that a human would normally do except do that at a pace and scale. That was 10 times, 100 times whatever the numbers are from a point of view of increase in scale. We also talked in that paper about how we saw novel attacks as well. These are the examples where you have frontier-based-style model capability stringing together vulnerabilities in a way that humans would never previously have been able to do—or to do at any pace or speed. Those things, luckily, were in the minority, but I think they're obviously a window into our future. But right now we see the pace and scale of just regular cyberattacks that have been AI enabled. So, from a national security perspective, I think it's absolutely imperative that we build AI capability in a defensive context for national security purposes. Mr HASTIE: Would you agree the global AI race is between the United States and China? Mr Anderson : I'd say that, if you look more broadly, they tend to be the two racing countries at the moment. Mr HASTIE: Sure. Are there risks with adopting open-weight Chinese models? Mr Anderson : I think we have to separate the idea of provenance and the inference aspect of it. I think there's an element where, if you're using a Chinese model that's hosted in China on Chinese infrastructure and you're sending your data there, that's very different to taking one of those models and hosting it in your own infrastructure or in a cloud provider's infrastructure, where there's not that connection. That's not to say that doesn't come with risk, because obviously you don't know how that model was trained, how it would feed you the information, how it will answer your questions et cetera. It's a very different national security risk. Mr HASTIE: In terms of building resilience for Australia, if we want to build a sovereign capability, we're obviously going to have to invite US frontier labs and hyperscalers to invest in our country. There's a lot of deep relationship with the United States built up over time at the strategic level. Mr Anderson : Yes. Mr HASTIE: It makes sense to partner with the US if we're going to partner with anyone. Mr Anderson : I would say so. We're investing not just from a technology perspective but also from a people perspective. A couple of years ago now, we invested significantly in putting bodies on the ground for what we call our MSTIC team, our Microsoft Threat Intelligence Center. These are the folks that are focused on nation state hunting, watching their tactics and techniques. We invested in actually putting one of those teams in country, and it can then liaise closely with the appropriate security clearances and also the ability to understand the national security landscape in Australia. So, for us, it's about the people and the technology as well. Mr HASTIE: Sure. Ms Fitch : Adding to Mr Anderson's remarks, I think the US-Australia alliance gives us a really strong foundation, and looking at multilateral relationships and alliances, including the Five Eyes, gives us a huge capability not only to build a digital backbone of security across those nations but also to consider our defence industrial base and economic opportunities as well. It's an area where we think not only that there's the imperative to branch out and have that security across allies but also that it could be a real economic opportunity for Australia. Mr HASTIE: For an Australian voice in the development of safeguards, is the Five Eyes forum the best place to exercise leverage and influence on some of these hyperscalers like Microsoft and frontier labs like Anthropic and OpenAI? Ms Fitch : It's a great question. I might ask Ms Craig to add in terms of AI safety. I think where we're seeing partnerships with the safety institutes across some of these countries is a really great place to start in terms of safety. I know, in terms of national security and defence, there are some additional considerations there, but I might throw to Ms Craig. Ms Craig : I would absolutely agree that the AI safety and security institutes and the network that has been developing over the last couple of years are one place that we see has a really important role in building capacity for all of the frontier governance work that governments are increasingly investing in. We see that AI safety and security institutes around the world are developing the capacity to evaluate frontier AI capabilities. They're developing the capacity to understand risk management practices and really are sharing learnings and even conducting things like joint evaluations. So this is absolutely a place where we see an opportunity to continue to invest from an Australian perspective with the recent creation of Australia's AI Safety Institute. This is an incredibly important capacity at the domestic level for AI governance but also at the international level for AI cooperation. Mr HASTIE: I have one last question, just to bring you back to where the chair started, with the opportunity cost of Australia missing out on building our own sovereign capability and potentially partnering with Microsoft and other businesses out of the US. What's the opportunity cost in national security terms if we miss this moment, this opportunity? Mr Anderson : I would say the advancements and the proliferation of AI being used in an adversarial context. Our recent Digital defense report, as I mentioned a moment ago, speaks to that significantly. We absolutely need AI for defensive capabilities, otherwise we will be operating at human speed against an adversary that's operating at machine speed. For us, it's extraordinarily important that we take advantage of it right now. Mr HASTIE: Thank you. CHAIR: Senator Pocock. Senator DAVID POCOCK: Thank you for your time here this afternoon. Looking at the latest ATO Corporate tax transparency report, across the various Microsoft entities Microsoft paid $1.39 in tax per $100 of income. Is 1.39 per cent a fair return for Australians? Ms Fitch : Across our entities, we had a total contribution of $168 million; that was in terms of our income tax across our various entities. I'm aware of the tax office report you're referring to. Microsoft complies fully with Australian tax law. Taking your point on fairness for Australians and spreading the benefits, that corporate income tax is only one kind of tax we pay across our entities. We also make a series of other investments into Australia and the community. It's something we take seriously. Senator DAVID POCOCK: You pay payroll tax and those sorts of things that every big business pays here? Ms Fitch : We pay $168 million of income tax across our entities. We pay an additional amount of tax in those other tax measures you've just referred to. Senator DAVID POCOCK: If Aussies are paying 25 to 40-plus per cent tax, look at Microsoft, who are coming here, making a lot of money in Australia and then paying 1.39 per cent tax. Now we're hearing we need to welcome you to Australia to invest in data centres and these sorts of things. What is the return for Australians? I get that they're the current rules, but is that a fair return? It just doesn't stack up for me. Ms Fitch : In terms of spreading the benefits, this year Microsoft invested $25 billion in data centre build-out. We believe in Australia's opportunity at this point in time, in terms of the work the government is doing on making sure Australians directly capture the benefit from those data centres and making sure they flow to community but also in terms of the investments that we're making in skilling three million Australians by the end of 2028, making sure that we are building an ecosystem that's investing in not just infrastructure but also skills, security and jobs. We had EY do some analysis earlier this year which, for the financial year you're referring to, found that we had a $36 billion economic contribution to Australia and that our presence here supported over 180,000 jobs. We're very conscious of our tax contributions and complying with Australian tax law, but we're also conscious of building an ecosystem in Australia where, structurally, those benefits are going to flow to Australians. Senator DAVID POCOCK: Your $168 million of tax—what was your revenue here? Ms Fitch : I'll have to take that on notice. Senator DAVID POCOCK: How come you can tell me your tax paid but not your revenue? Ms Fitch : I didn't bring the revenue number with me, I'm sorry. I'll have to take that on notice. Senator DAVID POCOCK: One of the other things we're being told is that these data centres are going to be great for the country and that there's a huge build-out; you said $25 billion from Microsoft alone. Looking at Microsoft's Australian data centre business, it generated $2.3 billion in income during the 2024-25 financial year but reported zero in taxable income and income tax payable. What is the long-term benefit going to be given? You have the big construction costs but you then find ways of basically reducing your taxable income, and then we're hearing that, potentially every five to 10 years, you have to replace the chips and then you do that again. What is the long-term benefit if you're not paying tax? Ms Fitch : I think the tax rule you're referring to there was developed during COVID to allow for business to continue in that infrastructure investment. As a result, the significant capital expenditure we've made during a period of time has allowed Microsoft to continue investing in Australia. We are very committed to making sure that we continue our economic uplift in Australia. Senator DAVID POCOCK: I saw you'd done deals with Australian unions. Ms Fitch : Yes. Senator DAVID POCOCK: What's that worth? Ms Fitch : You're going to have to clarify what you mean, sorry. Senator DAVID POCOCK: What sort of financial payment or contribution—or what's that deal worth? Ms Fitch : It's not a deal; I wouldn't characterise it as a deal. In January of this year, we were really proud to sign an agreement with the Council of Trade Unions. That's built on really listening to the worker voice, looking at things like skilling, looking at things like shared advocacy, and really trying to find a way forward together on some of these big issues around AI. We were able to have a workers summit in April, where we had 13 unions and 20 delegates, so that we could have our executives really understand and learn from that worker voice. We're partnering with the Trade Union Institute around union skilling—making sure that union delegates are being skilled around AI and that they're setting up those two-way information flows, to make sure that we're not only providing workers with the information that they need but also we're really receiving that worker voice and input on what best practice consultation looks like. Some of the takeaways have been absolutely fantastic, in the way that some consultations, in best practice workplaces, they have shared so willingly, and, where things have gone wrong, they have shared so willingly as well. I think unions have a really critical voice in this. We feel very fortunate to have such an agreement with the ACTU. I wouldn't characterise that as a deal. Senator DAVID POCOCK: Sure. That's great to know. In terms of the investment that Microsoft is making: in October 2024, the PM announced a $5 billion investment, and then, in April this year, you announced an MOU with the Australian government and a $25 billion investment. Is that $30 billion in total, or is the $5 billion part of the $25 billion? What does that $25 billion do? Ms Fitch : Just to correct the record slightly there, it was October 2023. The $25 billion this year is that additional contribution. It's a mix of capital and operational expenditure. I will ask Mr Black to add to that, in terms of that investment. Mr Black : Yes. The 2023 investment was $5 billion to expand our Australian data centre footprint, but also to invest in skilling and cybersecurity. It was actually the announcement where we also committed to the Microsoft-ASD Cyber Shield initiative, which is about threat intelligence sharing. With respect to the $5 billion specifically, that was an expansion of Microsoft's data centre footprint in Australia from 20 to 29 data centre sites, which was fully delivered over that two-year period. The $25 billion is an additional investment out to 2029. As Ms Fitch said, that's a mixture of capital and operational expenditure and it is intended to expand our compute capacity by over 140 per cent over that period. Senator DAVID POCOCK: Finally, on copyright: the New York Times is currently taking OpenAI and Microsoft to court, and a whole bunch of documents have been released as part of it. A Microsoft director, Brent Hecht, is quoted as calling the AI build-out 'the largest theft of labor in human history', and he privately stated that applying a fair use defence to AI data scraping would 'make a complete mockery of the concept of fair use'. Does Microsoft see Australia's copyright laws as a barrier to AI training data centres in Australia? Ms Fitch : I will ask Mr Gimbel to answer that. Mr Gimbel : Hopefully you can hear me. First, I'd like to say that I am familiar with that statement, but, second, that that statement doesn't reflect an opinion of the company or a statement about the copyright laws. What it does reflect is that these conversations are dynamic. There are multiple views. And I think it also reflects the nature of these types of technological developments. When these technological developments come online, when they come about, they always generate a bunch of different conversations. On the court cases that you mentioned, in the US: as you mentioned—and as has been talked about earlier—those court cases have slowly been working through the system. Some of those court cases have been already decided. In two cases that were decided last summer, in the Northern District here in California, it was decided that the process of collecting data and preparing that data and using that data for training materials for AI models is a fair use. So we're slowly seeing these cases develop. As you know, we are also a party to some of these cases, such as the New York Times case that you correctly identified there. The last part of your question asked about Australia and Australian law. I think you've heard from other parties about the current state of Australian law. Moreover, you've also heard from the Attorney-General on the application of the current state of Australian law as it relates to the process of gathering materials, tokenising those materials and performing a computational analysis in furtherance of the AI training models. Senator DAVID POCOCK: Mr Gimbel, my question was: does Microsoft see our copyright laws as a barrier to having AI training and data centres here in Australia? Mr Gimbel : We see your copyright laws as providing a high level of uncertainty about the ability to engage in the training processes that are necessary to train a model. Senator DAVID POCOCK: What's uncertain about them? They've been the same for decades, and, in many different realms, copyright deals get struck with collecting institutions. It's very well set up for that. What's so uncertain about it all? Mr Gimbel : The uncertainty comes into play in terms of the data that's collected, prepared, tokenised and used in that computational analysis step for training. What's uncertain is whether that particular copy qualifies as an exception. That's the uncertain part. You've heard from the Attorney-General about the uncertainty about that. Senator DAVID POCOCK: Sure. So the uncertainty is around the largest theft of labour in human history and whether there's a workaround to that in our laws or not. Mr Gimbel : No, that's a mischaracterisation of what I said. I said the uncertainty is around the computational analysis step that occurs in the process of training these AI models—again, the process of collecting these massive amounts of data, cleaning and processing them, translating them into tokens and numerical representations and then performing a computational analysis on the token, extracting patterns and facts and information to train the parameters of these models. Senator DAVID POCOCK: With 'collecting data', you make it all sound very dry, but this is people's copyrighted work. This is people's livelihood. Here in Australia, we have those copyright laws. So, in plain terms, our copyright laws would prohibit you from using collected—potentially stolen—works here in Australia. That's kind of what I'm hearing. I don't want to put words in your mouth. Mr Gimbel : First, I appreciate this question. Microsoft is a large copyright owner. We have been through the history of our 50-plus years, when it comes to Xbox Game Studios and our software. At the same time, I'll tell you a personal story. My wife is a documentary filmmaker and a creator within the industry at large, so I get these questions and I appreciate the nature of them. I think that—to get back to your question here—yes, the copyright laws in Australia don't provide the certainty the model developers need, which you've heard about in terms of that collection process and that computational analysis step that's required in the training. Senator DAVID POCOCK: Thank you. CHAIR: I'll move on to Senator Sheldon. Senator SHELDON: Ms Fitch, back in May, Microsoft's area vice-president for Australia and New Zealand, Steven Miller, wrote an op-ed in the Financial Review titled 'Microsoft wants workers at the centre of AI adoption'. It spoke about how the introduction of AI at workplaces will only work where employers are 'embedding workers' voices throughout the adoption lifecycle, from design and procurement through to deployment and review'. Can you just expand on what that actually means in practical terms in a workplace? Ms Fitch : Absolutely. I think what we're seeing across our customer base and across the economy is that not only are workers able to trust technology more when they are able to have visibility over how it is being implemented but the adoption rate—and continued adoption rate—is much higher and employers also have much better results. I might ask Mr Black to add further if that's okay. Mr Black : We've been part of Minister Rishworth's process around this as well, in terms of developing guidance for best-practice adoption of AI in workplaces, which we think is an excellent initiative. As part of that, I know the minister's consulting with employers, with unions and with other representatives. From our perspective, we think that worker engagement and consultation is not just a tick-a-box exercise; it is very much fundamental to successful adoption in the workplace. That's what we see in how we're adopting it internally within Microsoft, in one of the lessons that we have learned, but also what we're seeing across our customer base here in Australia. If we think about AI deployment and adoption, this isn't just an IT upgrade. We see it fundamentally as an organisational and cultural change process, and, as part of that, we think about what makes those sorts of processes successful in any other context. That includes visible leadership, two-way communication and early engagement and proper consultation and involvement of your workforce. So we see it as fundamentally reinforcing the opportunities of successful adoption, which will garner the productivity benefits that we see from this technology as well. Senator SHELDON: The work health and safety laws in the country already include the requirement for workers to be consulted. They have a mechanism of broad guidelines, specific industry guidelines and guidelines for companies. It's a system that works up and down at the same time. When companies have a new practice, they add to that conversation all the way up, and vice versa. There is a considered industry-wide response that people work within. Fundamentally, it works with the workforce and the employer, or the representatives of both the workforce and the company managers, making decisions about how to best implement safe practices. One of the things in the evidence that has been coming forward, as well as in the submissions that have been made to this inquiry, is about the psychosocial hazards that are happening with work intensification—issues related to things such as rostering, where there isn't human decision-making. Human intelligence isn't being used. The technology is being used without that intervention or oversight, or understanding, compassion or empathy. As a way of resolving issues and in light of the frame you just pointed out to me then, after my last question, is the OH&S framework in its broadest context a sensible way to approach the introduction of AI in workplaces? Mr Black : We certainly think that that's an important part of the consideration as part of all of the existing laws that apply to the way in which AI should be adopted, including antidiscrimination, privacy and other law as well. Our general principle around the regulation of AI is to think about that existing foundation that we have and ensure that it's fit for purpose for AI. There are examples of that in privacy, where we are very supportive of reform to ensure that those regulations are updated to reflect the environment which we are in now. I might also ask Ms Craig to elaborate here with respect to the sensitive-use case approach that we have within Microsoft that considers these sorts of issues in terms of the sensitive uses within workplace contexts as well, Senator—if that would be helpful—and the way in which we think about it. Ms Craig : I might even take a step back and share a little bit about how Microsoft approaches the set of issues that you're referring to in the development of our technology and the deployment of the technology, including in the context of workplaces. In the development of the technology, I might even broaden out beyond the workplace and say that the risks that you mentioned around psychosocial issues in interacting with this technology are something that we are taking really seriously in how we are developing AI technologies, so that, whether the interactions are with users in the workplace or in other contexts, the technology is performing in a way that is really reducing those psychosocial risks. Just in the past year, we've actually refined our policies in this area. We recently—last month—published our Responsible AI transparency report that talks a little bit about our policies in this space, and in this latest version what's called the Sensitive Interactions Policy, where we're really thinking about how we can from the start, in developing this technology, have clear policies for our product teams around psychosocial risk areas. How the technology is interacting with users in what we call these sensitive interactions, where there is this risk of emotional entanglement, or there is a risk, for example, of overreliance on unqualified AI advice and how AI technology is responding to users in those kinds of interactions and providing unqualified advice in a way that is not clear is a risk that we want to mitigate, and we want to test for. We want to mitigate it in the way that we are developing the technology and, hopefully, avoid some of the scenarios that you're talking about in the use, the workplace and otherwise. So we've developed and implemented these policies. We've actually also invested in industry-wide development of ways to measure how these policies are working in practice. One of the mechanisms is benchmarks that are really able to measure or evaluate the performance of the technology and show how different AI technology products are comparing on those psychosocial risk areas. We've been co-chairing an industry standards effort to develop a benchmark that really does that work in collaboration with others in industry and civil society. Ultimately, it's really important for the deployment of this technology that we have those kinds of benchmarks to be able to communicate transparently and publicly about the performance on those risk areas— Senator SHELDON: Maybe— Ms Craig : Yes, I can leave it there. I didn't get to the abuses, but yes. Senator SHELDON: By all means, put other information you want to share with us on notice. You're saying there's something Microsoft's done. I'd be particularly interested in having that documentation sent over to me. Also, how far off do you feel is that consensus in industry that you think is being created? Can you provide a rough timeline. CHAIR: I'm conscious of time also. Senator SHELDON: Yes. You might take that on notice, and I'll just ask you this one question then. An agreement was reached, Ms Fitch, announced with Nine Entertainment to licence their work. I don't think that your competitors have had similar local agreements. So are you ahead of the curve, and why not the others? Mr Black : I might pass to Mr Gimbel to add to this as well. That particular agreement is one that we are proud of. It reflects the kinds of activities that we have, not just here in Australia but also across the globe, with respect to copyright, to demonstrate the opportunities to work and partner with rights holders. In this respect, it was specifically in relation to grounding of Copilot results. I might see if— Senator SHELDON: I might go to the point then: why not others? Why haven't there been agreements reached with others? Mr Black : I might pass to Mr Gimbel to add to that, just in terms of enterprise, and some context around the broader activity that we've done in this space, and where this sits in the context of a lot of the pilot approaches that we have been rolling out. Mr Gimbel : Exactly. There have been deals with others. It might not be with Australian IP owners, but there have been global deals with others. In particular, similar to the Nine deal, in terms of bringing content and premium content to the grounding experience or bringing premium content to support the delivery of our AI tools and services, in particular with our Copilot Daily news in the UK and the US, we worked with large publishing houses to package up their content as part of a daily feed that would go to subscribers of that content. So those are examples of deals. Another scenario that I think is relevant here in this question is that we're developing a model, a marketplace for these types of arrangements. We refer to that as the Publisher Content Marketplace, PCM. We're happy to send you some materials on that. We just announced that. But that's us. Again, to my colleague's point about piloting different models that might work in this space in terms of providing a demand and supply side for the marketplace, the demand side are AI tools and services. The supply side are premium content providers like Nine media or others, and we've rolled that out in different jurisdictions. It hasn't been rolled out in Australia. But in those jurisdictions, we are piloting this idea to bring this content to the grounding phase of AI tools and services. Those are just a few examples, but I think that shows you our position, that we do believe that different uses of content payment can be an appropriate mechanism. CHAIR: I'm conscious that we are over time. I understand Mr Violi did have a question. I wonder, Mr Violi, whether you might be able to ask it and perhaps ask for it on notice. Mr VIOLI: I want to go to nuclear energy. I know you're part of the World Nuclear Association. You have partnerships with the US, Canada, Poland and the United Kingdom. I'm assuming those agreements stack up financially and from a safety perspective. If we could have a few minutes, could you explain those agreements. Can you confirm that they stack up commercially and from a safety perspective? If they stack up in all those countries, why aren't you looking at nuclear in Australia given its zero emissions baseload power and that data centres require power 24/7? CHAIR: I am conscious that we are already over time. I'll ask that they take a response to that on notice and come back to the committee with that. So I will conclude the session now. If you have been asked to provide responses on notice, could you do so to the secretariat within the next two weeks. Thank you very much for your contributions..