CALABRESE, Mr Sam, Director, Communications and Campaigns, Minderoo Foundation [by audio link] [15:11] CHAIR: Welcome. I understand that information on parliamentary privilege and the protection of witnesses giving evidence to this inquiry has been provided for you. I now invite you to make a short opening statement. Following that, we can get underway with some questions from the committee. Over to you. Mr Calabrese : Thank you, Chair and members of the committee. I appreciate the opportunity to contribute today. I look forward to discussing what our research is telling us Australian's feel and think about AI and the policy issues we've been working on at Minderoo, including safeguards and copyright. Minderoo Foundation is an Australian philanthropy. Our work on AI and technology policy started in 2019, when we identified an opportunity to bring together experts from the United States and China to attract to dialogue to help build a shared understanding around some of the highest risk military uses of AI, including the importance of meaningful human control. Contrasting with this work to help prevent harm to people, we also fund partners who use artificial intelligence and technology to help improve people's lives, including using AI to improve outcomes for brain cancer patients by helping match them to the treatments that most likely work for them. Working on both sides of the coin—preventing harms and unlocking the benefits—has given us an appreciation for the need to land balanced AI policy. In 2025, Minderoo convened SafeAI to support this goal. SafeAI brings together research and evidence, technical and policy expertise, civil society and decision-makers, with a focus on ensuring Australians can capture the benefits of AI safely. A key part of SafeAI's work is a comprehensive national research program that tracks what Australians think about AI. We have conducted three nationally representative surveys and a variety of focus groups over the past year, with the reports published on our website. The consistent finding is that Australians are not opposed to the technology. Rather, they can see the potential benefits in areas such as productivity and health care, but they also want confidence that the risks are understood and being managed. Our latest survey shows that people are becoming more familiar with AI and they're increasingly using it in their everyday lives, but that confidence is not keeping pace. Concern about the risk is growing and support for stronger safeguards is increasing. Half of those surveyed now think the risk of AI outweighs its benefits, and 70 per cent now believe AI is moving too fast. Despite this, the majority, 52 per cent, of Australians still prefer a balanced approach to AI governance, with clear rules that protect people while allowing innovation. The risk for those seeking to capture the potential benefits is that, when a balanced approach is taken off the table and people are asked to choose between only minimal rules or stronger safeguards, a clear majority of people choose stronger safeguards. This shows the biggest risk to innovation and productivity isn't the technology but public trust. Without confidence that someone has their hands on the steering wheel, Australians will forfeit benefits in favour of protection. In terms of what can be done to build public confidence, our research shows immediate harms remain at the centre of concerns. These are the challenges Australians see and can understand now, like AI generated mis- and disinformation and deepfakes, AI being used to supercharge cybercrime, personal data privacy and surveillance. Our submission sets out a number of practical actions we believe can help address these concerns. Given the committee's interest in copyright and AI training, I'll quickly conclude with a high-level overview of the work Minderoo is doing in this space and what Australians are saying in our research. For 25 years Minderoo has proudly supported Australia's creative sector, including the arts, journalism and public interest media. The discussions around copyright and AI model training bring together our support for Australian creators with our advocacy to ensure AI is safe and in the best interests of Australians. It is clear the current system is not working. Australian creative work is being used to train AI models largely without permission or payment. Because this training occurs outside of Australia, under the laws of other jurisdictions, Australian creators face an uncertain and costly path to recourse. With AI companies looking to build critical infrastructure in Australia because of our natural advantages and stability, there's a window of opportunity to address this unfairness now, while we can get both parties to the table. To support finding a balanced solution, Minderoo has commissioned economic, legal and policy work to examine a broad range of potential frameworks that would meet expectations of fairness for Australian creators and give responsible AI developers clarity to invest and operate here. Attracting that investment, particularly in AI model training, would bring these activities within the reach of Australian law, ensuring companies are held to our expectations, including fair compensation for rights owners and creators. It also presents an economic opportunity, with our preliminary modelling estimating that AI training alone could add about $7 billion a year to Australia's GDP by 2035. This is on top of anticipated inference and productivity gains. Just as importantly, by having the infrastructure here and building our sovereign capability, Australia has access to more safety levers and greater influence over how AI is developed and deployed, rather than inheriting the rules from elsewhere. Our research shows Australians are concerned about this issue, but they're open to a fair and workable compromise that ensures creators are paid when their work is used. Without addressing the status quo, Australian creators will continue to go uncompensated, with limited control over their work. Without a workable copyright framework covering AI training, investments may go elsewhere, and with them the greater influence Australia could otherwise have over the safeguards and systems we use. Consultation on our copyright report is ongoing, but we would welcome the opportunity for the experts who have contributed to the work to brief committee members at a later time. Thank you. CHAIR: Thank you, Mr Calabrese, for those opening remarks. It's really interesting to talk through the public trust issue. I think that's front of mind for a number of Australians. One of the main things that I see is that we could get all of the settings right, but if there isn't a sense of trust, confidence in the technology and confidence in our ability to keep people safe, the adoption isn't there. From your perspective, what is the most effective way that trust can be improved? Do you see that connection in terms of trust and uptake and, therefore, the ability to truly capture the opportunities that this technology provides for us? Mr Calabrese : Our research through both our surveys and the work we do with focus groups has really highlighted, as I mentioned in the speech, that the default position of Australians is a desire to find some sort of balance. They do not believe that there is a need to trade off safety for innovation. The concerns they're feeling, as I mentioned, are in relation to those really here-and-now issues like deepfakes, privacy issues and supercharging criminal misuse. They're the sort of issues that are causing people to have doubt. That doubt then fuels this desire to see stricter laws and stricter safeguards put in place. As I mentioned, when that balance is removed, when people feel like they don't have an option that includes both safety and innovation, they look to lock things down, they look to protect themselves, and that drives support for strict regulations at the expense of innovation. There are a number of solutions that we test around that. For example, we've discussed, with our focus groups, the AI Safety Institute and the government's National AI Plan. We found that when initiatives like this are explained to people, there is a large amount of support for them; the majority of people support the intention behind these initiatives, for argument's sake. In addition to that, we explore and test things around the ongoing privacy reforms at the moment, things that are actively being done to try to improve and protect personal privacy and data privacy. They're things that people are supporting that are building trust. They are visible actions that are underway right now. In addition to that, there are some issues that we have been advocating for, like greater recourse for everyday Australians to take when the Australian jurisdiction. CHAIR: Moving to the copyright issue, that is clearly a live issue. We've got a sense that it's not working for creatives now. As you mentioned in your opening remarks, and this is the evidence we have consistently heard, creatives do not have control now, they do not have an opportunity to be properly compensated now. There is a potential, if we can use this opportunity, to get a better framework in order to deliver greater control and compensation for creatives as a means to also enable training here in Australia. But, quite rightly, there are conversations around the extent to which—the way that may be described is we are undermining the rights holders and creatives by seeking to find a way to enable training as well as securing a better outcome for creatives. I know that you have mentioned there is work ongoing, and you've used this opportunity to provide an additional briefing to this committee on this work that you're doing, but is there a way in which you can provide an update to the committee on what you see are the current red lines and ways forward that we may be able to genuinely secure a better outcome for creatives and, at the same time, enable Australia and Australians to gain the benefits that we're talking about here when it comes to the ability to train? Mr Calabrese : Absolutely; I'm very happy to. We've been working with BDO Australia, Clayton Utz and Perfect Consulting—they are experts in the copyright field both here and internationally—to try to understand what options are available, what the challenges are and to put together a framework that finds the right balance. For us, that balance is finding a solution that supports direct licensing, supports collective licencing and, importantly, provides the safety net that essentially would look after everyday Australian creatives who, right now, are being opted into a system that doesn't provide them with compensation, that doesn't request their permission to use their material. At a high level, that's what we've been trying to identify, working through it with these experts, and it has presented a number of options. An option that we prefer, and that we would like to continue consulting on, is a limited statutory permission. Essentially, that works by ensuring that AI training is conducted on lawfully accessed material, and it requires an opt-out system—again, providing those everyday Australians who have been included in a system that they've not agreed to the ability to opt out. For those collecting societies and organisations with the resources and capacity to directly negotiate, it will allow them to do so directly with the AI developers; for those who don't have the capacity—often referred to as the long tail, which could include everyday Australians who contribute their thoughts and expertise on website blogs or otherwise—not to leave them behind, by ensuring that, as part of a permission system—this statutory permission—there's also a remuneration fund that these companies need to pay into; and, for those Australians who can be identified to have contributed work to AI training, that they are remunerated for that work. Then, more broadly, being able to ensure that, as part of that, there is some sort of consideration given to a creative industries fund or the like that benefits the Australian creative sector. CHAIR: Deputy Chair. Senator DUNIAM: Nothing from me, thank you. CHAIR: Senator Pocock. Senator DAVID POCOCK: Thank you for your submission and the work you're doing in this space. You've clearly done some work on a potential way to deal with copyright, and your submission talks about the need for regulatory clarity on copyright in model training. I was just wondering what is unclear about the copyright system, as it has operated now for decades. Mr Calabrese : During our consultation, some of the concerns that were raised with us around this need for clarity were presented across sectors and across industry. Industry that we were engaging with would talk about how the AI model training currently happening in the United States uses the fair use provisions to lawfully access that material, and the clarity on whether accessing that material through those US laws would apply here or otherwise. For the creative sector as well, there were questions around their material being used for training in the US, outside of Australian law, and the opportunities for them to opt out and withdraw that material or be compensated for it. There were a variety of questions that we were receiving. The goal there, or what we were trying to express, was that we would love to be able to give everyone involved absolute clarity around what the intentions are and, maybe, what the issues are that we need to resolve and the like. Senator DAVID POCOCK: Do you have any financial relationships with any of the big labs? Mr Calabrese : No. Senator DAVID POCOCK: In terms of regulatory clarity, one of the concerns raised by the head of the copyright agency is that, if you have this workaround copyright and opt-out system—which SBS and others have said is fairly meaningless because they still get crawled, and we know that there's this grey market for datasets—how do you stop Australia from becoming a global data-laundering hotspot? Mr Calabrese : Thank you for the question. The work that we've been doing specifically looks at AI model training. We've carved out things like inference and output. We believe that current copyright laws need to be applied to inference and output. There's definitely clarity that can be provided on that front in terms of expectations that those laws apply. Just because output comes from an AI model does not mean that it circumvents or is exempted from existing law. Senator DAVID POCOCK: You aren't concerned that, if we provide a workaround to copyright, all of a sudden, we're going to have, potentially, open slather here and that, if these court cases in the US do find that some of this training is not, indeed, fair use, we've essentially created a workaround meaning that Australian creatives won't be able to get any of that compensation. Mr Calabrese : There are a number of factors that were considered as part of the work we're doing, and I'm very happy to make the experts that have been involved in this report and this modelling available to brief you on those specifics. But where we've landed is a system that we believe fits within existing laws, whether Australian or international. In terms of your question about additional training occurring in Australia, the view we have is that right now every day Australians aren't benefiting. They don't have control over their work and they're not being fairly remunerated. That's something that needs to be addressed. The system and the model that we believe is most likely to help improve that allows the direct negotiations where required, as I mentioned, as well as a statutory permission or a safety net that captures value to ensure that everyone involved, everyone who's produced content involved in AI model training, is remunerated for doing so, particularly those who don't have the means or the capability to directly negotiate with AI companies or others. Senator DAVID POCOCK: Finally, looking at your submission, you emphasise how AI escalates gender based violence, impersonation and child exploitation. You endorse an approach that is safety by design. Do you think we should be banning things like deepfakes in our elections and potentially looking at ways to ensure that any Australian, if there is a deepfake of them, has an avenue for that to be taken down? Mr Calabrese : What we can see in our research is that issues like deepfakes are incredibly concerning to Australians, and they do want to see action in regard to them. We work with a number of partners looking into areas like this, including domestically with Digital Rights Watch, and they would probably have the expertise to provide a more detailed response in terms of a policy position. But what I would say is we can see that issues like deep fakes and mis- and disinformation are definitely top-of-the-list concerns for Australians. CHAIR: Senator Sheldon. Senator SHELDON: Thanks for joining us this afternoon, Mr Calabrese. I'm just wondering whether, in the research that you've done, you have got some concrete or practical policy work on how data centres might operate in Australia in a sustainable way with the support of local communities. What are some of the elements that that research may have drawn out? Mr Calabrese : This is an issue that we have been tracking over time as well, specifically attitudes towards data centre policies. In our most recent research, we can see that there's strong support for national standards for AI companies and ensuring that there are legal requirements for them to follow in addition to other policies such as ensuring that water use is minimised and that energy consumption doesn't increase costs to households. There is also an expectation that the topic of data centres isn't something that's left only to local government in a planning aspect but that all levels of government will be involved. We are working with domestic partners like the Human Technology Institute, and they've been doing a lot of policy work, leading the way on what these minimum data centre standards would look like. I'm sure that they would be happy to go into a bit more detail on that, but, suffice to say, Australians do want to see a higher level of standards and transparency applied to data centres. Senator SHELDON: I'm sorry. I just missed it. You were saying that someone may have a bit more information which would be of assistance to the committee. Who were you mentioning? I missed it. We could pursue that as well. Mr Calabrese : That was the Human Technology Institute at the University of Technology Sydney. Senator SHELDON: Thank you. I guess there are a couple of things. There's a view on where the data centres should be situated, and you've given a broad outline of some expectations of community on where a data centre might be situated. Are there any broader views about locations of building compute capacity where it might be geographically placed, or is it purely 'If it meets these criteria, then it may be okay to be in my local area'? Mr Calabrese : The most relevant discussion we've had with people about this would be through some of our recent focus groups. The responses were very high level, not specific to a particular area, but broadly people felt uncomfortable when they felt data centres were being built in their backyard, being built in their neighbourhood or being built in areas that would otherwise provide an opportunity to build more housing supply. But attached to that—I've mentioned some of the concerns and expectations around setting these higher level standards—there was also support in general for building data centres in Australia. As I mentioned in my opening comments, data privacy issues around data security are front-of-mind issues for people right now. What they can see or believe is that, by building the infrastructure here, Australia is going to have more control over that infrastructure and the data in those data centres here rather than that being elsewhere around the world. So there was an undertone of strong support for building sovereign capability and our sovereign capacity through data centres. Senator SHELDON: Just on the research you've done so far—I know you've mentioned there's been ongoing research and similar questions—is the intention to keep continuing to ask the questions on attitudes to AI into the future as well? Mr Calabrese : Yes, absolutely. We will both ask and provide that information to the public so that policymakers, industry, everyone can see how moods or sentiment may be adjusting over time. That's what we're finding really helpful, obviously, with a tracked poll. We're not looking at just one specific point in time. We're able to see how the issue evolves and feelings towards it evolve based on what's happening in people's lives and in society more broadly. CHAIR: Excellent. Thank you. I don't think there are any further questions. Thank you very much for your time and the evidence that you have provided us today, Mr Calabrese. I really appreciate it. Please engage directly with the secretariat to deliver any answers to questions taken on notice. Thanks very much. .