ALLEN, Dr Darcy, Private capacity GRUNDY, Ms Emily, Policy Officer, Good Ancestors Policy LANE, Dr Aaron, Private capacity SADLER, Mr Greg, Chief Executive Officer, Good Ancestors Policy CHAIR: Welcome. I understand that information on parliamentary privilege and the protection of witnesses giving evidence to parliamentary committees has been provided to you. Is someone going to make some opening comments? Mr Sadler : Thanks for inviting us to give evidence today. Good Ancestors wants to help Australia's leaders be good ancestors to future generations by making forward-looking and evidence based policy recommendations. We think artificial intelligence could be a force for good and a gift that we give to future generations, but it could also be readily misused and cause harm. You're probably already familiar with some of these risks. 'Hi Mum' scams use AI to clone a child's voice to trick parents into sending money to scammers. Only a few years ago, the idea that AI could clone a child's voice well enough to fool their own parents sounded like science fiction. Now anyone can do it for only a few cents. To understand where this technology is headed and what that means for the future of crime, we need to understand what it's already capable of. Last year, an AI model ran an automated cyberespionage campaign targeting governments and critical infrastructure. AI did 80 to 90 per cent of the work autonomously, with human direction only at key moments. Cutting-edge risks like this will become widespread and new risks will emerge. Anthropic's new AI model, Mythos, is another example. A 'zero day' is an undiscovered cyber vulnerability. They're the crown jewels of cybersecurity. Eighteen months ago, an AI model discovered a zero day for the first time. Now, Mythos has found thousands in all major operating systems and browsers—findings independently confirmed by the UK's AI Security Institute. Frontier capabilities rapidly become widespread. That could upend the cybersecurity landscape for everyday Australians and cause real problems even for businesses and governments that have invested in cybersecurity. This is an obvious problem in the cyber domain but could be even more worrying in areas where criminals have previously had limited capabilities. AI could give a wider range of criminals capabilities, from building biological weapons to conducting election interference. We think it's unfair and unrealistic to expect everyday Australians or even sophisticated businesses to cope with these risks alone. So what we're recommending is a multilayered defence. First, we have to understand the frontier. We need transparency from AI companies about the risks that they're creating and when things are going wrong. We also need government capacity to understand and manage those risks. Currently, the disclosures that AI companies make and that we depend on to understand these risks are largely voluntary. Second, we need to stop the worst of these capabilities from falling into the wrong hands. We should require AI companies to test their models for dangerous capabilities and not make dangerous capabilities available without suitable safeguards. One study found that only three per cent of the hundreds of biological AI tools surveyed had any safeguards at all. Finally, Australia's defenders need access to more capable AI than attackers can obtain on the open market. Neither the Australian government nor any Australian business was given early access to Mythos to harden critical systems like Medicare or the power grid. AI is moving so fast that we can't take a wait-and-see approach. Highly capable AI is here now and has real implications, from cyber- to bioterrorism. I'm happy to take questions. CHAIR: Thank you for those comments. We will receive and table these opening comments. Does anyone want to add anything? Dr Allen : Yes, we have opening statements as well. Good morning. I'm here with my colleague Dr Aaron Lane. We are scholars in economics and law from RMIT University in Melbourne. I am also a director of the Digital Economy Council of Australia, and Aaron has practical legal experience as a member of the Victorian Bar. Together, we have spent over a decade researching emerging technologies, including cryptocurrencies, AI and digital platforms. Our submission has two main themes. First, we describe how criminals behave like entrepreneurs. Like productive entrepreneurs, they spot opportunities, allocate resources, experiment with new technologies and adapt to changing circumstances. Indeed, cryptocurrency is one technology that has shifted the economics of crime in complex ways. But policy interventions can change the form of criminal activity rather than simply eliminate it. Policymakers should consider how efforts to suppress one avenue of criminal activity may redirect activity into other areas. The second part of our submission provides direct empirical evidence from Australian courts. The number of criminal cases involving cryptocurrency in Australia has grown significantly, from a handful a decade ago to 119 cases in 2024. Courts are already dealing with fraud, theft, drug trafficking and money laundering involving cryptocurrency. In most of these cases, offenders were caught using traditional investigative methods, such as physical surveillance, intercepted packages and admissions during questioning. This tells us that many offenders are adopting cryptocurrency not because they are highly sophisticated but because it lowers costs or is simply the most convenient option available to them. More broadly, our framework extends beyond cryptocurrencies. Other technologies, including AI models, can also act as a technological shock to the opportunity space for both unproductive criminal behaviour and productive entrepreneurial behaviour. Effective responses to this will require not just public enforcement but also new types of private governance. We welcome any questions that the committee may have regarding our submission. Thank you. CHAIR: Thank you very much for your submissions and, particularly, for making yourself available today. It is a very different world to what it was even 10 years ago, let alone three decades ago. Criminals have access to a lot more resources and money. It's very evident they're buying services that are far more sophisticated. We're particularly interested to hear your views about where we go from where we are now, because I think we're at a point where we have to now get ahead of the game. Your contributions as evidence will be very welcome. I'm going to hand over to Mr Batt as deputy chair to commence with some questions. Mr BATT: To Dr Lane and Dr Allen, to start with: in your submission you talk about crypto crime understood as a taxonomy—I'm trying to get my head around what all that means—and why different technologies enable different kinds of crime and why different regulatory responses are needed for each. Can you break that down a bit, and, also, how would you propose crypto crime taxonomy assist law enforcement in practice? Dr Lane : I might deal with the first part of the question and hand over to Darcy for the second part. The taxonomy that we've developed comes out of this idea that there's a tendency to treat crypto crime as a single entity and that everything using crypto is the same thing. We have tried, with our research, to show that there are multiple different categories. We've made a distinction between two methods, theft and fraud—so directly stealing a cryptocurrency wallet or cryptocurrency keys that give you access to the wallet versus fraud, which is deceiving somebody into providing that to you. That's a well-known distinction in the criminal law. We then have different contexts that that occurs in. We have pretty conventional fraud, which is seen as investment scams and romance scams. Those sorts of things have pre-dated cryptocurrency, yet the scammers in those spaces have adopted this technology as a payment mechanism, which helps get funds out of the country very quickly. But there's nothing particularly new about an investment scam. We then see intermediary enabled crime, where you are stealing not from an individual but from a cryptocurrency exchange that might hold consumer assets. We're then seeing activity in decentralised environments as well. The point of all this is: how does it assist law enforcement? It focuses the mind on: 'Where is the main game? What is the main type of offending that we're seeing? What are the most significant harms?' There's no point thinking about the worst crimes imaginable in a hypothetical sense; it's more about, 'We've got to deal with the real world, we don't have unlimited resources and we want to target enforcement into where the real harm is happening.' That's how we think the framework can be a useful starting point. That's not to say it's going to capture everything. I might hand over to Darcy, to add to that. Dr Allen : In crypto specifically, if we're talking about cryptocurrencies and digital assets, we have seen legislation passed this year in Australia that's being implemented at the moment around regulating and licensing centralised cryptocurrency exchanges. That's a fantastic move because it's showing we can't regulate the entire crypto economy, if we're talking about policy; there is a whole bunch of the crypto economy that doesn't exist within Australia's borders and it's not clear that any reasonable law enforcement could do anything about it. On top of Aaron's point, that means we should focus on those specific points at which we can have the most effective interventions—and in the crypto environment, that is those centralised exchanges. Broadly, often we get mixed up between the policy debates about how we respond to a frontier tech and what we should do in terms of enforcement. A technology comes along and it's very easy for us to imagine a whole bunch of harms around that technology, but it's much harder for us to understand all the potential benefits of that technology. We've seen crypto legislation after about 10 years of thinking about it. We're very early in the AI space and we're seeing it move very rapidly. I think it is having and putting resources into more adaptive enforcement and looking at crimes in practice rather than trying to predict where the technology is going to go. Mr BATT: Your submission implies that poorly designed interventions can make things harder for law enforcement by pushing criminals towards harder-to-detect methods. Have you got any examples, in an Australian context, of how that's happened, and have you got any solutions on how to fix that? Dr Lane : The place to start is thinking about intermediary enabled crime. If we regulate and say, 'We can't use cryptocurrency in Australia at all,' the point is not that people are not going to use it; to Darcy's point, it will strip out a lot of the benefits from that. It's not a hypothetical; we have seen jurisdictions around the world say: 'We're just going to flat-out ban it. We don't understand it. We're going to ban this technology until we can make it safe.' Similar conversations are happening at the moment on AI: 'There are all these harms. We need to stop development until we can make it safe.' It can push people out of a space where there's a beneficial economic exchange. If you're in a country that has banned this technology, all of a sudden you've pushed people in the crypto industry into a grey or dark market, and, rather than servicing real everyday consumers, they're now having to pivot their business model and start servicing money laundering, other crime syndicates, drug dealing and so on. That's the risk of that approach. Mr BATT: To Good Ancestors Policy: you recommend an AI act modelled partly on the EU AI Act. There are some criticisms about the EU act being a bit slow to implement, being complex et cetera. What lessons can Australia draw from the EU experience to avoid those pitfalls? Ms Grundy : In our submission we recommended an AI act. Importantly, we were recommending targeted AI specific legislation to fill the gaps that existing regulators are not already filling. You might have seen that the National AI Plan was published at the end of last year. In that the government signalled it wasn't pursuing a whole-of-economy AI act—in part perhaps it seems too complex, or perhaps it might overreach. There is quite a bit of criticism about the EU AI Act. We think that's reasonable and we think that, with many AI risks, existing regulators are fit to handle those risks in those domains. We think there are some specific risks which existing regulators aren't well placed to handle, and that's where we think it warrants new AI specific legislation. To provide a couple of examples—you can see these examples reflected in the EU AI Act and in places like California and New York, and even in emerging economies like Vietnam—the first thing we would recommend is around transparency. At the moment, AI companies provide some information about their models and the risks they carry, but Australia doesn't require them to provide this information. Other places like the EU and California require this information, but sometimes the companies just provide it voluntarily. This is critical information—which the government, as far as we're aware, is not requiring—about what risks the frontier models might carry. The second aspect is around incident reporting. What we would recommend is requiring incident reporting of critical safety incidents when they occur as a result of these models. Again, the EU AI Act requires this; California and New York also require this. That involves a couple of things. It involves setting a threshold of what classifies as a critical safety incident. This is something which could be loss of life or major harm; it's not your routine faults. It also involves setting a notification period: how long does the developer have to notify government of this? And it involves confidentially reporting this to the respective government. I think that's a key thing, because, even though other places like EU and California do require this, it's required to be confidentially reported to their government, not to ours. We think there are a number of lessons we could draw from the EU AI Act. In particular, the parts of legislation that target the biggest companies, the biggest risks and the most powerful models, particularly for transparency and incident reporting, are what we think is key. Mr BATT: Thank you. CHAIR: Senator Cox. Senator COX: Can I go back to Dr Allen and Dr Lane. I'll follow on from the deputy chair's questions in relation to the technology-neutral regulation that you talked about in your submission. You've recommended technology-neutral regulation, and I think you started to unpack that a little bit, but I'm really interested to understand if you have a practical example of how that might work. Dr Allen : The challenge when any new technology comes along is that we think we understand what the technology is but that the technology evolves very, very rapidly. I'll take cryptocurrency as an example. We were talking about regulating cryptocurrency from maybe 2016 or 2017. If at that time we had attempted to think about cryptocurrency, stablecoins didn't exist, DAOs didn't exist, NFTs didn't exist, and DeFi didn't exist in borrowing and lending. There's this huge area of the crypto economy that's been built with this technology that simply could not have been defined in any technical sense in legislation. I think we're going to see some of those challenges as well with AI. The pace of it is moving so quickly that we expect the applications and the technology itself will continue to move very quickly. When you have that rapidly moving environment, we need to think about how we can enforce against crimes in practice, and the way we should do that is we should look to existing laws to try and solve those, take them and go through the courts. I'm not sure if Aaron wants to add to this. Dr Lane : I think there are two things to say. The first is that, as a previous speaker pointed out, the government's response to the Senate Select Committee on Adopting Artificial Intelligence (AI) has been, 'No, we're not going to pass an AI-specific act, but we're going to go down the path of getting regulators and departments to look at their legislation and see whether there's anything in particular that might need updating.' I was certainly involved in the Treasury's look at consumer law, as one example, on AI. In the criminal space, we've got pretty robust criminal statutes, both the Crimes Act and the criminal procedure acts, that will allow law enforcement to bring action if there are perceived harms. I think that everything that has been described to me and that I've seen tends to fit within an existing criminal statute framework and can be brought through the courts. What we've seen, at least through cryptocurrency, is that courts are more than capable of dealing with cryptocurrency evidence. We've seen that before courts and we've seen that before juries. There's nothing in my mind that says that courts and juries won't be able to do that in a criminal setting with artificial intelligence either. Senator COX: Can you give us any examples of other jurisdictions where this is being successfully done? Dr Lane : I think, in the common law world, what we've seen—again, with cryptocurrency as an example—is the Corporations Amendment (Digital Assets Framework) Act 2026, which was passed earlier this year, does not regulate blockchain technology and does not regulate tokenised digital assets. What it regulates is companies that are engaged in exchanging digital assets and Australian dollars, or digital assets and digital assets. In the policy response—and this is not just in Australia but throughout the common law world, whether it's the UK or the US—the focus is on the intermediaries; it's on businesses and people, and it doesn't seek to define the technology in and of itself. Senator COX: Just before you came along, we had some police jurisdictions give us some information, and take some questions on notice from us, around some of the blockchain tracing. The privacy element of that is the tools around privacy coins and mixing services that are specifically designed to defeat blockchain tracing. Given that criminals obviously will adopt these when pressured, what's the realistic long-term effectiveness of blockchain analytics as an enforcement tool? Dr Lane : It's a good question. Dr Allen : We have this process—and this is the basis of our submission—that criminals will continue to adapt. As you shut down various parts of the ecosystem, the criminals will adapt and will be pushed further and further away from our regulated system, and I expect that we would see that over time, which is a challenge. But we also have to regulate in the context of the reality of what we can achieve. It's not clear to me what a single jurisdiction in the world could do about decentralised mixing services. I don't know anywhere where that's been successfully regulated. Because the criminals are acting globally, these are decentralised networks that are effectively impossible to shut down. I do see some sort of evolution towards that. My worry is, if we don't regulate effectively the lawful uses of the technology itself, then we won't get the benefits from the technology. We will just end up with a technology that increasingly pushes towards being out of our jurisdiction. Dr Lane : I think the other point to make is that the tracing tools are getting better all the time. The sophistication of these tools is increasing, and there's a bit of an evolutionary game going on here. There's a mixing service, or a tumbling service, that tries to obfuscate the funds, and some of the tracing technology—and this is essentially proprietary technology; there are three or four main companies that are involved in this software—is getting better all the time in response to the changing services and the changing business models that are going on. In parallel to that, we've seen, at an international level, law enforcement. In the United States, for example, there are mixing services that have been on the sanctions list. We've had people who are allegedly running, or at least developing, the software being charged with offences. So there's this cat-and-mouse game going on. As Darcy said, that's kind of the point of our submission: we're living in this evolutionary, dynamic environment, and we'd expect that to continue. Dr Allen : And what is interesting about that is that a whole bunch of this is being developed in the private sector itself. It's obviously being used by law enforcement as well. This is what we saw in the early days of the internet too. There were a bunch of companies that were quite incentivised to figure out how to find fraud in payment platforms, for instance, because they wanted those platforms to be legally operating. So it's just interesting—that private dynamic, as well, which is important. Senator COX: That leads into my next question—it's almost like you can see into my question pack! There are strong economic incentives for private actors to invest in what is crime prevention, right? But the police submission suggests that private-sector cooperation is pretty much inadequate or inconsistent. What is your assessment of why that gap exists? And what would you recommend to absolutely close that gap? Dr Lane : I've got some observations from my involvement in representing victims of cryptocurrency fraud and scams. One of the challenges is that law enforcement are receiving thousands of requests in any given week about fraud and scams and those sorts of things. So, increasingly, people are turning to private lawyers—and, admittedly, I am one of them—that are using civil processes, through the courts, to try and track and trace and seek recovery for victims. Often, you will find that digital currency exchanges will not be terribly cooperative with you. I think the Australian based exchanges are, but some of the international exchanges just do not comply with court orders and don't respond to communications and so on. I think there are probably efforts in the civil law that can be improved, but, in terms of law enforcement, again, requests have to be made. There is still a lot of manual labour that goes into this sort of investigation, and I think state and territory police forces need to be adequately resourced to be able to do that job. CHAIR: Thank you very much. We are quickly running out of time, but I will put a question so that we've got something on the record for people who—and there are lots of them—are tuning in and watching this, no doubt, as they do parliament every day. As to cryptocurrency, the conversation this morning has all been looking at the downsides—at the criminal use of it. Can you just quickly outline where cryptocurrency is being used and all the advantages of it, so that we have a full picture for the committee. Dr Allen : Cryptocurrency is often talked about just as a way to send payments, and that is an obvious way that we could use this technology, but, fundamentally, this is a new technology that enables us to have new types of property rights on the internet, and what that means is that we can have new types of ownership on the internet; we can have new organisational types. Cryptocurrencies and digital assets power a whole range of frontier business models that are happening. This is a really important frontier part of the economy. It's not just happening for money—it's not just for money or for speculation. There are real business models out there, building real products—including decentralised physical infrastructure. We have crypto projects where there are tokens that enable people or incentivise people to build wireless infrastructure in the economy—so real-life wireless communications, running using crypto tokens. There is a whole range of different business models. Money is one aspect. Crime is one use of that. But it is a tiny proportion of the overall crypto economy. CHAIR: Well, that's very good. I think it was important to get that on the public record, so I appreciate that. I have a question in relation to the AI institute. Good Ancestors, you mentioned in your submission that we need to have an institute. Can you elaborate a bit further on that and also cover off who would be responsible for setting that up and taking oversight? Mr Sadler : Since we made that submission recommending the creation of an AI safety institute—and we've made that recommendation here but also in other contexts—government, in the National AI Plan late last year, went ahead and created an AI safety institute. We're glad to see that, and I think it became operational just in the last week. This is a great step. It has functions relating to technical work to test emerging AI models, it has functions relating to international engagement and building international AI norms and it has a role in supporting existing government agencies and regulators. Our key reservation here is just that the Australian AI Safety Institute seems to be the worst funded of any equivalent organisation in the world. It has funding of $7½ million a year, which is forecast to decrease as the budget goes forward, despite government acknowledging that the impact of AI is increasing. If we wanted to make the Australian AISI equivalent to the UK example, which is the world-leading example, there would have to be seven to 10 times more budget than we've currently allocated it, based on GDP or population. So it's a substantial increase to match. That remit I was talking about for the AISI, the Australian AI Safety Institute, is also perhaps the broadest in the world. So it's got that technical role. It's got the international engagement role and this role of supporting agencies. I don't think there's a solid count, but Australia has well over 100 agencies, regulators and departments that the AI Safety Institute is expected to support on $7½ million a year. So there's a significant mismatch between this exceedingly broad scope and this exceedingly small budget. Another point is that the AI Safety Institute is not a regulator—which is fine; it's appropriate that it's not a regulator. But it does mean that, absent the support of a regulator, it's missing a lot of the tools it would need to do its job. Emily was talking about the need for disclosure of information, like what capabilities models have. If they can't benefit from someone who can disclose, that makes it harder for them to do their work. Similarly, there's no obligation in Australia for accidents and incidents to be reported, so we're sort of blind about the extent of AI harms happening in Australia currently. The AI Safety Institute can't demand access to models to perform this testing. So, yes, we're heartened to see that the AI Safety Institute has been created. It's only been operating for a few weeks, so we don't want to be unfairly critical of it, but I think it's got a very hard job to do. It has a very large job to do on a very small budget, so we'd like to see that increased. CHAIR: I do appreciate that. Unfortunately, we're out of time, but there may be further questions that we'll put on notice to you. If there's anything else that you deem important after today's hearing, we need to have that in by 3 July. Thank you very much. We appreciate your time..