CARTER, Mrs Roxanne, Head of IP and Media Policy, Government Affairs and Public Policy, Google LOVETT, Ms Stefanee, Director, Government Affairs and Public Policy, Google Australia STAFFELL, Mr Simon, Director Policy Partnerships, Google DeepMind [by video link] [15:51] CHAIR: Welcome. I understand that information regarding parliamentary privilege and the protection of witnesses giving evidence has been provided to you. I now invite you to make a very short opening statement to enable us to have enough time to ask questions. You can provide anything further directly to the secretariat. Ms Lovett : Thank you, Chair and committee members. You have our written submission, so today I want to focus on what AI can mean for everyday Australians. The real test of AI is whether it makes life better for people across Australia, regardless of what they do and whether they live in cities, suburbs or regional towns. That's why we're also investing in digital skilling, helping tens of thousands of Australians learn practical AI skills for their everyday work. We are seeing this real-world impact of AI through our Digital Future Initiative—a $1 billion investment in Australian AI, infrastructure, research and partnerships. In regional WA, we're partnering with the Lions Eye Institute on AI screening to help prevent avoidable vision loss in remote First Nations communities. With CSIRO, our FireSat collaboration will detect bushfires from space when they're as small as a classroom, providing updates every 20 minutes. And through Electric Saul with Rewiring Australia, an AI adviser is putting personalised guidance into households to help families navigate energy choices and cut electricity bills. To unlock these benefits safely, we build responsibility into our technology from the ground up—guided by our AI principles, technical red-teaming and frontier safety commitments, while supporting the Australian AI Safety Institute. To sustain this momentum and ensure the benefits of the technology flow, Australia needs an enabling copyright framework—one that works for both creators and innovators. That means sensible settings that respect creators and keep Australia attractive for long-term AI investment, so local builders can create models that understand Australian voices, law and culture. Australia is well placed to lead our region in AI, and the window to do that is open now. With clear, balanced settings that keep people safe and give local builders certainty, Australia can shape the global conversation rather than follow it. We look forward to your questions. CHAIR: I'll extend on that last point. What will it look like for Australia to play a leading role in how this technology is governed, and how do we shape that? What will that mean in connection to our national sovereign capability, in your view? Ms Lovett : Simon, I'll hand over to you first on the model piece, and then I will follow up with a response about what that means for Australia. Mr Staffell : Thank you very much for the question. This is an incredibly important question for us, and, while we are very focused within Google and DeepMind on taking a really responsible approach in terms of how we're deploying the technologies, we recognise that this is something that we can't do in isolation or alone. It's very much a collaborative process with governments and with wider society, and Australia can clearly play an incredibly important and leading role here. Specifically for frontier AI governance, we're at a moment where technology is clearly advancing very, very rapidly. That means that we need to continue to invest in research, and there's lots of opportunity. We are partnering with organisations in Australia in terms of research and how we can advance the science of evaluation of models and how we can make sure that we have a diverse approach to things like benchmark setting and those kinds of areas. So that's a really important area to continue on. With a lot of these issues being very nascent and evolving, we also think there's a real need at the moment that we have a real focus on standardisation that's global, and so that we have interoperable global standards. Again, there's a really important role that Australia and the ASC in Australia can play in terms of that wider agenda around how to create standardised best practice—a real focus on companies like iGoogle and Other Labs in terms of evaluations, benchmarks and mitigation. But how do we create a standardised approach? We've been very upfront and vocal about calling for increased focus on standards, and we think Australia could play a really important role there. Ms Lovett : Bringing it back to Australia in terms of what the benefits are or how we see the broader sovereign capability, having had the benefit of listening to some other witnesses, I won't repeat everything that they've said, but this is thinking about what that ecosystem builds out and what it means for Australia. When we've built data centres in the US or when we're building them in other jurisdictions, what we see is what comes off the back of that. It's about bringing the logistics. It's about bringing a supply chain. It's about bringing all those other things that can help make Australia an advanced manufacturing nation again. It's about how we make sure that the ecosystem more broadly sits across both SMBs, from an adoption point of view, and workers, with the ability to adopt or adapt and learn new skills. It's also about what it means from an economic point of view in terms of new opportunities, new markets and new sectors. CHAIR: I asked the previous witness this question. What do you think the opportunity cost would be if we missed out on the opportunity to chart our own course in this space rather than have it potentially just happen to us or have decisions made by others elsewhere? Ms Lovett : When we're looking at investment, these are really long-term investments for us. They're 20- or 30-year processes. So what does that look like in the long-term? I would say that the risk or the opportunity cost is not just a short-term impact for us; it's what the impact would be over the longer term. CHAIR: Mr Violi. Mr VIOLI: You just mentioned 20- and 30-year investments. Obviously data centres power AI and they are a long-term investment. They also take a lot of energy and rely on 24/7 power access. Can you talk through Google's global strategy when it comes to investing in nuclear power? I know you've got investments and agreements in the US and Finland. Can you talk through that strategy, please? Ms Lovett : Mr Violi, I heard you ask that question of the former witness. I don't have the details of that. I'm very happy to take that on notice. I can talk to you more generally about our approach, but— Mr VIOLI: You invest in the US. You invest in Finland. I'm assuming that those agreements stack up for Google commercially. Ms Lovett : As I said, I'm very happy to take that question on notice and provide the committee with what our response is in relation to nuclear. I can talk more generally about our approach to energy and renewable energy, if that would be helpful. Mr VIOLI: We've got a moratorium here in Australia. I understand that. But, given that Google as an organisation is investing in nuclear energy all around the world, it would be good to understand why you're not in Australia. I understand the moratorium, but there's the commercial aspect of it and the zero emissions reliable power. Is that something you would look at here in Australia if the moratorium was removed? Ms Lovett : I think that's a hypothetical that I can't answer on the spot. You pointed to the moratorium and that's the reason why we haven't looked at that here. More generally, what I would say, though, is we take the idea of bringing our own energy very seriously, so when we're looking at these investments, we want to be good neighbours and we want to be good partners in the locations where we build. What that means is being net additive to the energy grid. We bring our own energy. We look at what that means in terms of the power that we would use. In the context of Australia, we see a real opportunity that would help to catalyse renewable energy and the market in Australia, so at a high level that's our focus and consideration for Australia. Mr VIOLI: Given the time, I'll hand back. But it would be good to get on notice the commercials around those global projects. CHAIR: Ms Chaney. Ms CHANEY: Thank you very much. I have a general one to start. We're hearing about the important role that Australia has to play. Global consistency—I'm interested in what role you see Australia playing. Obviously, the most consistent thing is the lowest common denominator across different jurisdictions when it comes to safety standards. But Australia potentially has a leadership role as well, and not just to comply with the lowest common denominator. How do you see Australia's role in regulating safety standards in order to drive increased trust, balancing that global consistency versus leadership roles? Mr Staffell : I would start with my previous point about standards. I think there is an opportunity at the moment to create this global consistency. There are areas where a number of countries are really focused on having AI safety institutes, including Australia. We think there's a huge opportunity in organisations like AI safety institutes to play a role in some of this nascent emerging standardisation in areas like benchmarks, ensuring that we have a diversity of testing frameworks and benchmarks. That's an area that an organisation like the AISI could play a leading role. There's also importance in the coordination between these AISIs. The network of AISIs and how they collaborate together is another area where the Australian AISI could potentially lean in and play a really important role as we create this global interoperable approach to testing frameworks and benchmarks and evaluations. There are also things potentially that could be learned from other AISIs as the Australian AISI is standing up. The UK AISI and the CAISI have really learned a lot of lessons and I think shown a lot of good practice in terms of building a really credible bench of experts, building the evaluation approaches and the frameworks for how you need to get that talent in AI governance issues within government, and that can have a wider catalytic effect in the technical capacity within government that actually extends beyond the safety issues and into the wider capacity of government to be able to adapt and adjust to these challenges as they're emerging. Ms CHANEY: At the moment, the AI Safety Institute in Australia has 25 employees across all its functions. What sort of scale do you think is necessary in order for it to be making a serious contribution to those global standards on testing and evaluation? Mr Staffell : I wouldn't say there is a specific number of people that the organisation needs to recruit. Clearly, you can do a lot with a small number of people. It's more about, I would say, the focus areas and the technical capacity of these individuals. It is really a handful of people that have that technical capacity globally, so if you had the right skill set and the focus on these niche areas of technical research, which can add real value and complement what's happening in other places, for example, in this network, you could add force multiplier effect beyond the small group of people that you have if they're really focused on areas that can add value to that wider picture and collaborations. Ms CHANEY: But do you think, as it's currently resourced, it's able to play that role, or is it still in the early days? Mr Staffell : I would say, just factually, it's in the early days. I believe it is relatively nascent. It's getting established and growing. I look forward to seeing how that continues to develop, and I look forward to working more with them as they develop and expand and build this capacity. Ms CHANEY: Critical incident reporting—I'm interested in whether you would be supportive of Australia having a mandatory critical incident monitoring and reporting framework, and, if so, what considerations should be taken into account in designing that framework. Mr Staffell : It's obviously a very topical issue at the moment. I'll perhaps give a view on the general approach here and then turn to my colleague Stef in terms of a specific Australian context. In terms of serious incidents and reporting, it's one of the areas where standardisation is important, which I've talked about. We do need to make progress on having standardised approaches to what the severity of the incident is and what the appropriate timeline is for reporting in a particular case of severity. There is an amount of that at the moment which is on the companies to make determinations and decisions. Clearly, if we are in a situation where we judge that there is an important incident, we would feel it's very important for us to do our piece to inform people and to notify. Transparency is incredibly important to our approach. But we do think there's an important piece at the moment around standardised best practice across the industry so there's a consistent framework for knowing what is severe, what the gradations of severity are and what timelines are appropriate to those levels of severity. That's one of the areas where standardisation could be really impactful. Ms CHANEY: Who is best placed to develop what a reasonable expectation is, in terms of critical incident reporting to governments? Is it AI safety institutes? Will the obligations be on the platforms? How do you see those standards actually being developed? Mr Staffell : Again, in the local Australian context, I would pass to my colleagues on the panel there. The general approach, I would say, is that there is an opportunity at the moment to have more in the space of global standards—working with standards organisations like ISO as well as working with a network of ASES to create a framework. A lot of this is currently with frontier labs making some of these decisions, and with companies like Google taking our obligations very seriously in terms of reporting and having those approaches in place. The companies working together in terms of standard-setting and collaboration could advance that, and working with standards bodies globally could create some of that standardised practice. Ms CHANEY: I have one more question. On biological weapon risk, I understand that there are big uplifts in capacity expected in the next generation of models. I'm interested in Google's perspective on that and what you think governments should be doing to prepare for or guard against that risk. Mr Staffell : It's a very important question and a very big focus for Google DeepMind, with our heritage working on science with AlphaFold, AlphaGenome and some of our science models and research. It really speaks to the inherent dual-use nature of some of these technologies. The incredible benefits that we're seeing across science and ensuring that our models, including our science models, are able to be used by scientists to take forward incredible areas of research in health does, at the same time, have this risk of these tools being used by malicious actors. Our approach starts with our frontier safety framework, which is a set of protocols to help us identify and mitigate potential misuse and potentially severe risk in areas like CBRN, or chemical, biological, radiological—including cyber as well. That's a set of protocols and internal frameworks where we identify, pre release, where there are risks and place mitigations around those risks in a range of different technical ways in terms of training our models. We then also put in place postrelease monitoring. An important part of that, clearly, is internal governance and what we do within the company and what our peers in other companies are really needing to focus on as well. Your question also points to this point around how we support the wider ecosystem and use of these technologies. We think that areas like supporting scientists to use our science models on areas like biosurveillance, tools that could potentially advance areas very beneficial to biosecurity, could really help to get the defence ecosystem ahead on areas like biorisk, which were not created by AI but are a big focus because of the advancing capabilities, using technology to advance some of those frameworks. Ms CHANEY: Given the size of the potential risk, do you think that it's reasonable for governments to be trusting internal protocols of companies, or are there mandatory requirements that should be imposed to ensure that there's a level playing field and all AI developers are dealing with these risks in a similar or consistent way? Mr Staffell : I think there's a huge amount of practice that's happening within companies, like Google and our peers, at the moment that's focused on some of these nascent areas and adaption. I would say that we're at a point at the moment where a lot of this could be moved more into standard setting across industry. We have a common framework and approach to how we approach some of these issues that take some of those frameworks and protocols that we have within the company and standardise them across the industry. I do think we're at a point at the moment where it's ripe for progress, and we have actually seen a lot of progress in terms of what companies are calling for in creating that consistency. Ms CHANEY: Beyond playing a part in global standards, how do you see the Australian government's role in protecting Australians? It could take a long time to develop global standards on some of these really significant risks. Should we in the meantime be doing what we can in domestic legislation to protect Australians while we wait for those global standards to develop? Mr Staffell : Stef, I might pass to you on domestic legislation. Ms Lovett : What I would say is that, as Simon talked to, this is a really important, fundamental and foundational issue. I think that we want to ensure that Australia has a seat at the table. What we wouldn't necessarily want to see is a framework where we're not able to work across a number of jurisdictions. We're very happy to engage with the government, we want to understand what the risks and concerns may be, we want to ensure that we continue to have an open dialogue on a number of these issues as we have, including on things that I was going to point towards before—around critical infrastructure reforms that have happened over the last couple of years—and we want to ensure that we can share expertise, knowledge and best practice of things that are happening, both here and overseas, to make sure that's incorporated into our laws here. Ms CHANEY: Thank you, Chair. CHAIR: Mr Watts. Mr WATTS: Obviously, Google DeepMind has a long history in the UK, but am I right in thinking that training for Gemini only happens in the US? Mrs Carter: Yes, at the moment we do the bulk of our training in the US under the fair-use doctrine. Mr WATTS: What are the kinds of factors that you would take into consideration when evaluating different jurisdictions for making an investment in the very substantial training clusters that would be necessary for doing training for Gemini? Ms Lovett : There are a range of factors. Part of that are trade-offs about the consideration of high-cost or low-cost jurisdictions. It's regulatory certainty or policy certainty on a range of issues. It's cost inputs, whether that's construction, acquisition of land or how we think about power and water—what the site looks like. It's considerations around local communities. I think it was pointed to earlier about whether those communities want a data centre in that location or not. As a good neighbour, that is critically important to us. On the regulatory piece or the stability, as I mentioned earlier, these are really long-term investments. It's probably important to think about them in two different buckets. The first is our data centres that we may use for our own operations. That's about serving our customers, whether they're consumers or whether they're enterprise customers. Separately are AI infrastructure hubs and what they might look like. The regulatory considerations are consistent across those two. Obviously, as we've talked about previously, copyright is one of those considerations. Mr WATTS: Those large-scale training clusters don't necessarily have to happen near end users. They could be in regional areas, away from large urban areas—is that right? Ms Lovett : That's correct. In the context of serving customers, you want to be close to subsea cable networks and connectivity so that there's low latency in the production of those services. That's not an immediate concern when it comes to training. So, yes, they can be in non-metro, more rural or regional areas. Mr WATTS: Okay. You've got the sort of spectrum of cost trade-off issues on land acquisition, construction costs, energy costs, all those kinds of things, and that's a spectrum that you can weigh up compared to other jurisdictions. But then you have these gating decisions—like the regulatory environment, like a reliable investment environment that you can make a multi-decade investment in. I think Australia ticks that. It's a trusted strategic environment that sensitive customers will be comfortable with. You can tick that for Australia. You've raised the copyright issue a couple of times here. Would you be able to articulate, one, what the red lines are that would cause concern about making a training investment in the context of copyright? And two, could you disentangle a little bit some of the different kinds of agreements that Google has made with content providers? The deals that Google has done have been cited by a number of rights holders as examples saying that everything's fine, actually; there's not a need for copyright reform. For example, there's the deal with the Guardian and the deal with Stack Overflow. Can you segment out a little bit the different purposes for which you'd do a deal with a content provider versus a deal for vectorisation of data for training? Mrs Carter: Sure. I'd like to start by just saying that ideally what we're trying to get to with copyright frameworks is that of course you want to protect a private interest but you also want to support a public good, which is the sharing of knowledge, future creativity and the ability to have innovation, research and scientific endeavours. Trying to achieve that balance is really, really critically important. When it comes to the responsible development of AI, we really need to train on a huge amount of content, a really diverse set of content, and we do talk about access to training on the internet as the lifeblood of AI. But if you think about the internet, the internet is sprawling and messy and there is no way to know what is copyright protected online. It's unclear what might fall into an 'orphan work' or what might be in the public domain. And it's not static. These things fall out; what might be copyright protected today may not be tomorrow. Also, we have to recognise that some content falls under exceptions as well. I listened to some of the sessions this morning, and I think there is a sense that we only need to train on high-quality content. We actually need all the content. We need the good. We need at times the bad and ugly. For example, if we are designing these tools to fight spam, we need to be able to train on spam. It really goes to the challenge that we have with Australia; it's not a question—I think, Senator Pocock, you were getting to this with Microsoft—of lack of clarity. It's the sense of an obligation to go out and get prior authorisation and licensing agreements for all the content that we would want to train on. If that is really the obligation, just a conservative estimate on the back of a napkin would show that to train a medium-sized large language model would take us around 200 years to be able to get the necessary authorisation—which then takes us to the second point. That is related to content that is freely available online. There is a market that is developing. There are deals being done, and that is for content that is behind paywalls, for example, or archived or off-platform. As Google, we have the heft, the architecture and the partnership agreements to go out and do those deals. We recognise that we really want to do as much of those as possible and build the scale, and so we launched relatively recently something called the 'content pilot offer', which is designed to help surface the more local niche content so that we can do those direct deals. Mr WATTS: People in that conception can opt-out of that training via technical measures like robots.txt instructions or a paywall that makes content publicly inaccessible. There are ways of doing that. We should note that these aren't new issues. Google has been dealing with these issues in Australia since the advent of Google Search—is that right? Which products has this come up for in the past for Google and resulted in different decisions about where to make investments in training, expertise and human capital? Mrs Carter: I would have to go back and take that one on notice. Again, I heard you speak earlier about the challenges that this may have posed for Google Search and Google Maps. I do think that generative AI and AI are slightly different because of the way that the models are learning. I don't want to get into the technical details and weeds of this, but I do think it's important to highlight that these are not retrieval systems. They are not designed to reproduce or retrieve information. There are no copies in the model. They are learning. It's computational analysis. It is the digital equivalent of reading, which is fundamentally different, I think, to what we may have considered in the past. Mr WATTS: And that's what vectorisation is as a technical term. Mrs Carter: Just finally on this point, Google has lots of options about where to make these investments, doesn't it? There's no imperative, from your perspective, that any change happens in Australia at all. There are lots of other places that you could be doing this, theoretically. Ms Lovett : As Roxanne mentioned earlier, we are training our models in the US. As an Australian citizen, I would love for us to be able to also have that infrastructure investment here, on the back of the sorts of benefits that we discussed earlier that would flow to our community but while also respecting the position of trying to find a middle ground for creators as well as AI developers. That's outside of just Google. That's also looking at Australian AI developers that might want to be able to develop models here. Mr WATTS: Thank you. CHAIR: Deputy Chair. Senator DUNIAM: Mr Watts's question covered off on a couple I was looking at myself, but— Mr WATTS: They were very good questions! Senator DUNIAM: They were excellent; I have to commend you on those. In Dario Amodei's essay where he talks about safeguarding and risk mitigation, the concept of embedded evaluators—is that something Google has a view on as something that might be worth pursuing here in Australia as a result of everything that happened in the last couple of months? Mr Staffell : We've long worked with external evaluators as part of our AI development process. Built into that process is a role for external evaluation. It's a key part of the Frontier Safety Framework, which I've mentioned: rigorous evaluation across model development, extensive internal red teaming, external evaluation from third-party specialists and research into the latest issues that are emerging and capabilities that are emerging. You'll have seen with the announcement of our Gemini 4, released last week, how we're further adapting our approaches because of some of the testing that we've done and therefore taken account and doubled down on even more of some of our safety protocols. Having evaluation and third parties as part of that process is really important. It's also important for that to be a diverse group of external partners, with specialised research in many cases, and to work with civil society and with AI safety institutes, which we talked about quite a bit already, all bringing in expertise to stress-test the models. We really look forward to expanding those efforts. Broadly, in working with evaluators, we think that's really important and that there is more we can do to continue to build on that framework and investment that we've made. Senator DUNIAM: Thinking about it in the context of social licence, it is one of the bridges we need to cross on the way to where I think most people want to get to. Copyright's another one, as is energy security and other standards. So a government sanctioned or government authored type of set-up is something Google would not oppose in a uniform approach to external evaluation across all the frontier labs and other entities? Mr Staffell : What I would say—and I've talked quite a bit about standards already—is that there is a role at the moment for maturation of that standards approach and making it consistent across. We've also long called for an approach to AI that's bold and responsible, balancing the opportunity with the potential risks. Technology is too important not to regulate but also too important not to regulate well. So that's the kind of operating framework that we stand to. We do also think that it's important to have an approach that isn't duplicative, which is why I've pointed to the role of international standards, and that we have a consistent approach that matures some of these nascent fields into an international approach that is standardised and has best practice so that we have all of the expertise that needs to come from across the field, across society and working with governments into a consistent, interoperable global framework. Senator DUNIAM: A final question on that point around non-duplicative efforts. Are there regulatory frameworks or arrangements internationally we should look to as a guide to what we might do here that have worked well? I know that there are variations. We've talked about one of them being copyright already. Or are there things we should completely avoid? Mr Staffell : Of course there's regulation that could be considered at different areas of the stack in terms of applications, in terms of models. If I speak to models and my colleagues may wish to speak to other parts to that. In terms of models, clearly there are a lot of emerging discussions happening in the US around different proposals. We are engaging very closely with that and think that there's potential to create some of the standardised practice in some of those areas that we've talked about in terms of benchmark setting, evaluations and reporting through some of those proposals in the US, in terms of legislation, which could create a consistent approach. CHAIR: There's just one clarifying question from Senator Dolega, and then I'll throw to Senator Pocock. Senator DOLEGA:   Mr Staffell, following on from the deputy chair's question, when you were talking earlier today about how you support the principle of independent evaluation. Earlier today, we've seen from the Anthropic evidence that they were more supportive of the independent evaluation where they have the ability to report without editorial control of Anthropic. Do you support that sort of openness when reviewing these particular models or emerging AI technologies? Mr Staffell : Your question speaks to transparency, which runs really throughout the whole field from publishing of what— Senator DOLEGA: And that editorial control of Google in particular. Mr Staffell : I'm sorry. Senator DOLEGA: Sorry—without editorial control of Google or without you controlling the narrative. An independent evaluation where they can report to the public without you controlling the narrative. Mr Staffell : This is one of the areas that I think it's really important that we make progress on a standardised approach. As I say, we work very closely with external evaluators at the moment in getting the right guardrails in place, privacy protections in place and security protections in place—which, of course, has been a high profile issue at the moment around how do we make sure that we have model security and the right kind of environments for them to do the testing in a secure way. We do think that there's a lot of good work happening in that space, but we need to continue to standardise some of that approach, including in the issues that you raised. CHAIR: Senator Pocock, noting that we've got a short time before we're due to finish. Senator DAVID POCOCK: Do you know the current value of tenders and contracts you hold with the Commonwealth government? Ms Lovett : I don't have a specific figure. I know we're working across a range of customers, but I'm happy to take that question on notice. Senator DAVID POCOCK: And maybe just on notice too, the value of the deed of standing offer with Defence. I'm still unclear about the long-term return to Australians. As an Australian citizen, I was looking at the ATO corporate tax transparency report. Firstly, congratulations. You have the highest tax rate of the big US tech companies. You pay 3.39 per cent tax on revenue. I'm interested if you think that is a fair rate of tax to be paying in this country. Ms Lovett : Senator, I would give you a similar response to what you received earlier, which is: we comply with the laws of Australia. Those accounts that we provide to the tax office are audited accounts. In addition to the tax that we are paying, we've been in Australia since 2002, and there are a number of initiatives and other investments that we make here. It's important for us to be a strong partner to the communities in which we operate, as I mentioned earlier, but it's not just about that tax that is paid. We've got roughly 2,000 people in Australia that work for us. We continue to build out and make further investments here, and we want to be here for the long term. Senator DAVID POCOCK: You talked earlier about the supply chain in building out data centres. My understanding is that 70 to 80 per cent of the data centre build is imported. Mainly, the cost is chips and other things. What is the other benefit of building these data centres in the long term? We have this sugar hit. We hear from the Reserve Bank that it's adding inflationary pressure at a time when households are being slugged. What can you say to Australians that, over the next 10 to 20 years, this is how you benefit from data centres? Ms Lovett : We might provide some additional information to the committee, but we made an investment announcement roughly this time last year in India. Off the back of that announcement, we have seen a number of corporations come in to build out their supply chain operations, logistics and other things around that investment. When I talk about what this means for the long term, it's about seeing the opportunity to have new industries, new manufacturing capability in Australia and long-term support for what that investment would be regardless of where that investment might be in terms of either a data centre that might be in a metro area or a data centre that might be in a regional location. In addition to that, it's about catalysing that renewable energy transformation in Australia, so it's looking at how do we help bring new clean energy and consideration around innovation or other things that would also support that data centre. That might be from a water perspective as well. Senator DAVID POCOCK: On that, Google's commitment is still net zero— Ms Lovett : By 2030. Senator DAVID POCOCK: Can you tell us about your training licence with Reddit? Mrs Carter: This is, again, a deal on access. As I was mentioning, we're looking at different deals that we can do with different content creators. What I can say is that these aren't simple transactional deals. We're meeting our partners where they are, so every deal is very, very different. We're trying to scale those, as I mentioned, on the content offer pilot. That is a bit of a challenge, again, because what we're trying to do is surface that more local, niche content and find the smaller content creators, but we also recognise that people want different things from these agreements. So that is very much a pilot. Senator DAVID POCOCK: So you're paying them to access Reddit content? Mrs Carter: We're paying to access the content for AI development, yes. Senator DAVID POCOCK: Why aren't you and others just engaging with copyright holders to strike these sorts of licensing agreements? To your claim that it'll take 200 years, clearly, there's a long tail, and maybe there's something that needs to be done there. But my understanding is that, to do deals on 80 per cent of copyrighted materials, it's actually fairly straightforward. Collecting institutions are well set-up; the mechanisms are in place. Why are we hearing from basically all the big US tech firms that on the one hand you're building the most incredible technology the world has ever seen, and on the other hand it's simply too hard to do some deals on copyright? Mrs Carter: To speak frankly, I think it's because copyright—there is no single register of who owns what copyright, for example. Senator DAVID POCOCK: Have you spoken to any collecting institutions? Mrs Carter: Yes. We do, and we partner with collecting societies. We've got longstanding relationships with collecting societies—for example, on YouTube. The challenge that we have is that there is no collecting society that represents all the different types of content that we need to train the models, and in some instances as well our experience has been that some collecting societies are not able to identify what they own in their repertoire, for example, or able to help manage things like the controls that we are also building out in partnership with web publishers, which perhaps takes me to my second point. It was something that came out this morning quite clearly. It is not just simply about deals or payment. We've heard very much from web publishers that they want control and agency, so we have been building out a suite of controls. We mentioned earlier that we launched something in 2023 called Google-Extended, which allows any web publisher to say, 'We do not want our content used for model training.' That is something that we fully respect. We continue to engage in the ecosystem—for example, in a forum called the Internet Engineering Task Force, which is really trying to wrestle with something, again, that we've heard from rights holders: that it's getting increasingly complex out there and that they're having to navigate too many different types of controls. And, here, we're really trying to simplify it as much as we possibly can so that, again, they have that sense of agency. Senator DAVID POCOCK: Sure, but I'm confused. Why should the onus be on every Australian to attach some sort of text file or some sort of message to every single piece of their copyrighted material that is online to stop US tech companies from hoovering it up, creating a product and potentially displacing workers? Can you not see that this is what's happening? I think there is this moral part of what is happening here that just isn't being talked about, where US tech companies—because you're big, because you're powerful and because you've got President Trump on your side—are dictating terms to every single Australian. Why is that acceptable? Mrs Carter: I would challenge that framing. When we're designing our tools, we're trying to make them as inclusive as we possibly can. We want our tools to be reflective of local culture that speaks to Australian users, and business partners as well. Obviously, we want to train on Australian content, but we also are very mindful that they should have choice and control, so we have gone out and we have educated as well as we can about the suite of controls that we have, and we know that they're used. We know that they're used regularly and often. I think it comes back to trying to balance, again, this idea that we want to protect a private interest whilst also allowing a public interest, which is the development of AI. And how do we strike that balance? If we have to move to an opt-in, where we have to go out and get prior authorisation from everyone, then I just can't see how AI development can happen. Senator DAVID POCOCK: Ms Carter, you keep saying public interest, but this is technology that's owned by the most powerful corporations in the history of the world. It's private interest. It's US tech companies that own this, that are monetising this and that are looking to replace jobs at scale to recoup the trillions of dollars that you've pumped into developing it. I find it hard to cop that the wealthiest companies in the world talk about their technology being the public interest. Has Google trained on Australian content? Mrs Carter: As I said at the beginning, we will train on content that is freely available online. Senator DAVID POCOCK: That's not my question. Have you trained on Australian content? Mrs Carter: If it's freely available online. Senator DAVID POCOCK: What does 'freely available online' mean? Mrs Carter: If it is not behind a paywall, if there are no technical measures to restrict access to it, if it is lawfully accessed and if it's not opted out—content that is freely available online—then, under fair use in the US, we will train on that content. Senator DAVID POCOCK: Thank you. CHAIR: Thank you very much. Can you please forward to the secretariat within the next two weeks any information that you've taken on notice to provide to the committee. Thank you very much for your contributions. We really appreciate it..