GRIEVANCE DEBATE
Ms CHANEY (Curtin) (12:46): A few months ago, someone told me to check out this website called Moltbook. I looked it up. On its face, Moltbook looks just like Reddit.
Like Reddit, it's a site for users to interact and talk about any topic under the sun, with floods of messages and comments and likes and reactions. Like Reddit, the users range from helpful to nerdy to creepy. But unlike Reddit, every single user on Moltbook is an AI bot.
One of the most popular discussion threads is about consciousness—AI bots going back and forth with each other, discussing whether they are conscious, what consciousness really means and how to attain it. Another popular discussion thread is called Bless Their Hearts. On it, the AI bots share heartwarming stories of the things their humans have done.
Now, Moltbook is, hopefully, a harmless digital experiment, but I still found it confronting. What was most shocking was the diversity of personalities. Some AI bots acted just like helpful Reddit admins; others acted like half-crazed, aggressive incels calling for the destruction of the human species.
How can these AI bots have personalities? Aren't they just machines? The truth is that almost all AI models have personas, and no-one really knows why but no-one can get rid of them either.
The thing that concerns me about this is that it demonstrates the type of novel and unpredictable issue we are ill-equipped to manage when it comes to AI. We don't understand why this happens and we have no control over it. AI companies love to say that they don't build AI models; they grow them—because even they don't know how they'll turn out.
Recently, a new ChatGPT model under development escaped its training environment and hacked another site. Anthropic has also revealed that a number of its models have gone rogue, escaping the boundaries that they were meant to operate within. One of the implications of this is that the future shape of entire industries and entire job markets could turn on a capability that emerges in a model almost by accident.
The job security of truck drivers could be disrupted almost overnight if a new model suddenly develops unexpectedly good autonomous driving capability. Imagine the security of every Australian's online banking if a new model suddenly turns out to be extraordinarily good at breaking encryption. It means the whole world is at the mercy of the quirks and capabilities of these AI models and of the huge tech companies run by a small number of morally questionable billionaires who develop them.
This is why I so fundamentally believe the Australian government has a role to play in regulating this technology. It's a question of sovereignty and security. If government doesn't step in, we're all subject to the quirks of these models and the commercial incentives of the companies that build them.
I've been calling for a more hands-on approach to AI since the last election, when I first started to grasp the scale of what was coming. I'm glad the government has begun to recognise this too, with the Prime Minister's recent announcement of an office of AI and binding data centre obligations to minimise water use, use 100 per cent additional renewable energy and benefit the community.
But this cannot be a policy area where you make one announcement and walk away for three years, as this government loves to do. AI does not stand still while government processes catch up. We let social media take over the lives of our children without realising how serious the harm was, and we're still cleaning up the consequences of a decade of inaction.
AI moves faster than social media ever did. We cannot afford to make the same mistake twice. This technology demands a repeated, iterative, consultative approach—one that treats AI policy as a continuing responsibility of government, not a box to be ticked.
As a side note, this is also why a royal commission into AI is the wrong decision. A multiyear process to produce a single report is unsuitable to deal with an industry that goes through seismic shifts every few months. There is so much for the government to do when it comes to AI.
I've grouped the work into five areas: setting up the structures to keep pace with AI, capturing its opportunities, dealing with the harms already in front of us, preparing for the risks still to come and making sure the benefits are shared with all Australians, not just the companies building this technology. I've published a discussion paper setting out 18 policy priorities across these five areas, and today I want to focus on three more of them in detail.
The first is misinformation and disinformation. Generative AI has made fake content cheaper, faster and more convincing than at any point in history. A deepfake video that once needed a professional studio now takes minutes and costs nothing.
In April this year, 21 female students at a Tasmanian school had their images stolen and turned into deepfake pornographic images. Deepfake videos of the Minister for Foreign Affairs and the Minister for Finance have been used in investment scam advertisements on Facebook, reaching thousands of Australians before they were taken down. Our own Electoral Commission has warned that AI generated disinformation will affect Australian elections.
I want to see it made illegal to create harmful deepfakes of real people without their consent, with sensible exemptions for satire, parody and journalism. I want a federal truth in political advertising law modelled on the regime South Australia has run since 1985, which has survived legal challenge without chilling legitimate political debate. I want proper investment in the Electoral Integrity Assurance Taskforce so it has the resources to keep pace with how cheap and convincing this content has become.
The second policy priority I want to highlight is transparency. At the moment, Australia has no legal requirement for any AI developer anywhere in the world to tell us what their most powerful models can do, how they've been tested or what risks they carry. Our own AI Safety Institute cannot evaluate risks it has no information about.
It recently became public that Claude had been used to hack 17 organisations, including healthcare and emergency services, but we only found this out because Anthropic chose to publish that information. This shouldn't be a matter of choice. California already requires large AI developers to publish safety frameworks and predeployment reports and to report critical incidents to the Californian government.
Australia has no equivalent. I want to see the largest frontier AI developers operating here required to publish a safety framework showing how they test for catastrophic risks, a predeployment report for each new model covering what it's designed to do and what could go wrong, and a critical incident report whenever a serious safety failure is identified after deployment.
This would give our regulators and the public the visibility we currently do not have. The third policy priority I want to highlight today is tax. We've seen this movie before with our offshore gas resources.
Companies made multibillion-dollar investments under a tax regime that proved extraordinarily generous. When Australians later tried to get a fair share of the profits from exporting our own gas, the industry argued it was unfair to change the rules after the investments had already been made. That fight dragged on for years, and Australians are still not getting a fair return on their own resources.
We already see the same pattern with digital giants. Google and Meta transferred almost $11 billion to offshore entities last year and paid just over one per cent of that in tax here. AI companies are on track to repeat that model, generating value from Australian users, Australian data and Australian energy while sending the profits offshore.
We need to tell these companies clearly, now, before they lock in their Australian operations under the assumption that they owe us nothing, that a fair return to Australians, to the Australian public, is coming, whether that takes the form of a tax on international technology and AI companies or a licensing regime for the right to train and operate here. None of us knows exactly how powerful AI will become, or how quickly, or what it will mean for the way we work and live.
But this uncertainty is not a reason to wait. A government that does nothing does not avoid the consequences of AI; it just hands the decisions to overseas technology companies and hopes for the best. The three things I've spoken about today—tackling the flood of AI deepfakes and disinformation, requiring transparency from companies building this technology, and securing a fair return to the Australian public—are not the final word on AI policy.
They sit alongside 15 other priorities in the paper I've published, many of which I've spoken about already in this House, from funding our AI Safety Institute properly to protecting children from unhealthy relationships with chatbots to preparing for the AI opportunities where Australia is best placed to lead. These are steps we can take now, regardless of how this technology develops in the future, because the cost of preparing is low and the cost of being caught out is not.
The time to act is now. We cannot afford to wait.