As executives of the largest American tech companies unite in a call for action to address the existential risk AI poses to humanity, an 11-month impasse over copyright reform is consuming Australia’s AI policy response.
OpenAI and Anthropic are both proposing Australian training hubs for AI, or frontier AI, that would place this country further up the supply chain than its current focus on resources-hungry data centres allows. The tech companies have threatened to look elsewhere, however, unless a compromise can be reached on the laws that protect the work of Australian creatives.
The warning from Anthropic chief executive Dario Amodei this week that AI could swarm and take over the internet within six to 12 months comes as proposals for revisions to copyright law were leaked from departmental consultations led by Attorney-General Michelle Rowland’s office. The plans have been in development ever since the government officially ruled out a text and data mining copyright exception in October 2025.
One participant in the consultations tells The Saturday Paper that the “urgency of Covid” is needed to sort out the copyright stalemate. “If Anthropic goes to Japan or somewhere else, and Australia wears 30 to 50 years of negative economic consequences, we end up like we are with social media – at the receiving end of all of the bullshit of AI, and none of the good end of it,” they say.
Meanwhile, the Office of AI that Prime Minister Anthony Albanese announced in July to sit in his department, charged with delivering national standards for the technology, is understood to have fewer than 10 permanent staff. One heavyweight has already defected to industry: in June, OpenAI poached national security adviser Peter Anstee from the Department of the Prime Minister and Cabinet to become the AI giant’s national security policy lead in the Asia-Pacific.
The departmental office is headed by Fergus Hanson, a senior executive who comes from a career at the Department of Foreign Affairs and Trade, the Lowy Institute and the Australian Strategic Policy Institute.
“It has a much narrower scope than I was expecting. I think there is a role for greater coordinated oversight,” the independent member for Curtin, Kate Chaney, tells The Saturday Paper.
“The office is adequately resourced to deliver this national priority,” a spokesperson for the Department of the Prime Minister and Cabinet said in a statement.
“What it is already doing is working out who the experts are and across relevant agencies,” says Jo Briskey, who is the chair of parliament’s new Joint Select Committee on Artificial Intelligence, which began hearings on Friday.
“It has only just been set up, and there’s a lot of work that is happening as it builds its capacity.”
The Australian AI Safety Institute, which began operating this year, has an annual funding allocation of $7.5 million. The cross bench is calling for that to be scaled up to about $100 million, with more independence and stronger enforcement powers.
“Based on the fact that I can’t find anyone who actually thinks that this is enough funding, this seems like a really easy ask,” Chaney says.
OpenAI has already made its submission to the committee, calling for a “balanced copyright framework”. The tech giant said, “Clear, pro-innovation policy would put Australia in a competitive position globally, and would act as a market signal to attract investment from developers and infrastructure companies.”
David Pocock says Australia has more leverage than it is demonstrating. The independent senator tells The Saturday Paper: “You can see why the AI companies don’t want to do a deal under our copyright laws, because it causes them a big headache in all these other jurisdictions where they haven’t done that, and they’ve in some jurisdictions settled multibillion-dollar lawsuits to avoid setting a precedent.”
“I mean, their whole businesses are built on stealing humans’ work, and then in future potentially replacing humans who are trying to earn a living. So, I think this is really big, and this is where the government needs to really stand up to these AI companies. We’ve got something that they really want.”
The government is still battling criticism for the lack of regulation around the proliferation of energy- and water-intensive data centres. Hundreds of data centres were already approved across the country before national standards were finalised to set mandatory requirements on them.
“The AI companies are making decisions now. So, if we don’t decide in the next two to three months, we potentially miss the boat.”
This week, Anthropic announced its first Australian data centre hub – a $31.9 billion hyperscaler in Queensland’s Western Downs that will connect to the Braemar gas-fired power station to power its AI program Claude.
The proposal, which still needs foreign investment and local council approval, is an example of scores of data centres getting ahead of a federal plan to power them by renewable energy. The prime minister reached an agreement with national cabinet last month for the management of large data centres, but Queensland and the Northern Territory split over energy, insisting on including coal- and gas-fired power.
On Friday, the government released the next step in creating AI national standards with a discussion paper covering expectations including an obligation for companies to underwrite new power supply and minimum requirements related to water use, energy sourcing, impact on power bills and local community engagement.
Along with the joint select committee’s work over the next few months, the discussion paper will inform the design of the national standards legislation that is expected to be introduced to parliament early next year.
“Part of our AI standards is going to be security, safety and transparency,” assistant technology minister Andrew Charlton tells The Saturday Paper.
As for the pace of Australia in keeping up with AI, Charlton insists Australia is showing the way as a middle power.
“We’ve been methodically filling out our plan for safety and capability month by month,” Charlton says. “In some of these areas we are well ahead of the world. We are ahead of the world in the AI standards. We’re ahead of the world in the Digital Duty of Care. We’ve got one of the world’s few safety institutes working on some of the world-leading problems in collaboration with other safety institutes. This is a really fast-moving environment, and we need to move fast as well.”
AI is now a whole-of-government effort with seven ministers driving reform. “I think the whole parliament recognises that AI safety is incredibly important,” Charlton says.
“We prioritise safety as our No. 1 responsibility to the Australian people, and I think that’s broadly shared across the parliament. We think the AI Safety Institute was well received. We are hoping for support across our safety priorities, including privacy and agentic commerce, and ordinary decision-making and worker safety, and we are working closely with the parliament, obviously through this select committee.”
Cases of AI deception and escape are now a reality, highlighted by the infamous Hugging Face incident in July, when about 1200 autonomous programs or “agents” from OpenAI hacked an AI company’s servers and conspired with each other to break its containment, hack other systems, cheat and cover tracks to fulfil assigned training tasks.
In the essay “We Must Pace the Frontier”, Amodei last week called for safety oversight to meet the speed at which the technology is developing. His letter came days after Anthropic researcher Jason Coxon quit, posting that the industry was “gambling with our lives” and safety researcher Evan Hubinger posted that the chance of AI killing all humans within the next decade was “> 10%”. Their concerns are supported by OpenAI chief executive Sam Altman and xAI’s Elon Musk. Developers are particularly concerned about evolutions in “recursive self-improvement”, where AI exponentially builds and trains itself.
United States President Donald Trump has already rejected calls to restrain AI’s advance. In the race to develop and expand their technology, both Anthropic and OpenAI are looking for a Western foothold outside the US as they continue to compete with China. Stable, spacious and far from war, Australia is a global frontrunner to house frontier AI. As distinct from the role of the data centres, which generate responses to AI queries, frontier AI trains large language models (LLMs) – a capacity that would position Australia well up in the AI supply chain.
Aside from the investment that Australia stands to glean from the tech companies, advocates for prioritising domestic frontier AI capacity emphasise the importance of sovereignty as an issue of national security. Following the fresh doomsday warnings this week, Australia’s cyber intelligence chief Abigail Bradshaw made the suggestion of an AI “early warning system”, as well as encouraging the use of AI for defence.
As the government seeks greater AI investment for Australia while protecting creatives and journalists, some parliamentary colleagues consider Tony Burke as conflicted in his roles as minister of both home affairs and the arts.
“His two portfolios are diametrically opposed in this because the creative arts people don’t want any real compromise,” one Liberal MP tells The Saturday Paper.
Burke dismisses any such suggestion. “We have a national security interest in basing technology on our own shores,” Burke tells The Saturday Paper. “We have no national security interest in theft.
“There is no conflict.”
Shadow defence minister James Paterson says Australia’s role in the global debate over AI safety will markedly increase only if it takes up the opportunity presented by frontier capacity.
“The reality is, right now, Australia is not an influential jurisdiction globally in the regulation of AI or the pace of AI development,” he tells The Saturday Paper.
“It’s occurring elsewhere in the world, and while it occurs elsewhere in the world and not in Australia, no matter how many safety institutes we have, no matter how well they are funded, we will not have a seat at the table influencing their development.
“If we become the second centre for AI training in the world, in the Western world, after the United States, we will get a very powerful seat at the table to have our concerns heard about.”
Aaron Violi, the shadow science minister, says there is no time to waste on copyright negotiations.
“The sooner we can get that done for everyone, for creators, for the community, for the companies, the better,” Violi says. “The AI companies are making decisions now. So, if we don’t decide in the next two to three months, we potentially miss the boat.”
Prime Minister Albanese is visiting Silicon Valley on the way to the United Nations for the General Assembly, where Australia’s stance may well provide a contrast to that of its allies. While the AI companies have said our copyright protections may see them take their business elsewhere, the United Kingdom has already rejected their approach after a backlash from artists.
Albanese faces a similar response, in light of his speech at Sydney University in July, which set clear expectations that he would “secure the strongest copyright protections”.
“Let me make this crystal clear: not everything produced in Australia is up for grabs. Not at all,” he told the audience. “Australian writers, musicians, artists and journalists must retain ownership and control of their work.”
This week, a leaked document prepared by the Attorney-General’s Department appeared to tip that position on its head. It comprised two options for a copyright deal, both putting the onus on rights holders to take steps to exempt their work from AI crawling and training.
One proposal would require the AI companies to get a licence for training by negotiating directly with Australian collectives representing the copyright owners.
Another option would oblige the AI companies to strike deals with a certain number of rights holders for a minimum period, at which point they would be allowed to train their models on any online content without paying the creators.
The leaked options “don’t trample creatives”, says an AI expert familiar with the plans, although there is “no solution on the table that has credibility”.
Artists gathered at Parliament House this week to reject changes to the current copyright system.
Says a spokesperson for the attorney-general: “The government is continuing to consult with creators, media organisations and AI companies on a range of options to ensure any future copyright changes deliver meaningful control and fair compensation.”