A cluster of ministers and executives gathered in Canberra as the federal government unveiled its new national AI strategy, while climate and consumer groups massed outside to protest. The package, released on June 6, 2026, prioritises public spending and new institutions but leaves several complex regulatory choices unresolved. The announcement arrives as Microsoft, Anthropic and other AI firms intensify efforts to find stable, low-cost places to train large models and data-centre investment in Australia rose more than 60 percent to $2.6 billion in the September 2025 quarter. Campaigners warn that decisions made now will determine whether Australia captures long-term economic value or inherits a large emissions and safety bill.

Inside the Canberra meeting room, government ministers framed the strategy as a push to build capacity and keep Australia competitive, while outside interest groups put pressure on tougher rules. The government emphasised direct spending measures and the creation of new institutions, including an Artificial Intelligence Safety Institute, as the centrepiece of its approach. Industry figures in the room also heard the offer of incentives, even as critics said the package sidestepped the hardest regulatory questions about model training, liability and cross-border data flows.

Spending and institutions, not hard rules

Officials describe the strategy as pragmatic. The plan focuses on funding for research, grants to accelerate AI adoption in government and industry, and building institutional capacity to oversee emerging risks. Tim Ayres, the minister for Industry and Science, has defended the record by pointing to the new institutions and recent laws limiting social media access for under-16s. But commentators who reviewed the package say it's 'big-spending' on support and institutions while short on specific legal and liability rules that would govern how models are trained and used, according to a published critique.

That gap matters because several former officials and technologists argue voluntary standards and capacity building won't be enough to manage systemic risks. In a Canberra interview on Four Corners, former industry minister Ed Husic said he had been preparing standalone legislation with mandatory guardrails for high-risk AI before he was removed from the portfolio. Husic told the program, "We put AI regulation in a 'too hard' basket. We blinked in the face of Donald Trump and decided that we just couldn't get away with our own approach on this." His account suggests the choice to defer tougher rules was both political and strategic.

Australian officials now face a practical trade-off. Major AI companies, led by Microsoft and Anthropic, are actively seeking stable, low-cost locations to run training workloads, and Australia has become a major destination for new data-centre investment. Australian Bureau of Statistics data cited in reporting shows data-centre related investment rose more than 60 percent to $2.6 billion in the September 2025 quarter. That inflow promises jobs and local spending, but it also raises two linked concerns: energy demand and the social harms possible from deployed models.

Mastheads

Greenpeace Australia Pacific commissioned a report that uses Australian Energy Market Operator projections to model future electricity demand. The report warns that data centres could double from roughly three to six percent of national electricity demand by 2040 in a low-growth scenario. Under a higher-growth case, that share could climb much further, from about two to 13 percent. Greenpeace singled out the planned 1 gigawatt Mamre Road Data Centre Campus in Western Sydney and estimated the facility would produce about 1.3 million tonnes of CO2-equivalent emissions in 2032, equivalent at peak to 560,000 petrol cars.

Greenpeace's head of climate and energy said Australia is 'completely unprepared for the magnitude of impacts' from the planned buildout and urged a moratorium on new approvals.

Proponents of hosting training infrastructure argue the economic prize is large and immediate. One opinion piece urged Canberra to negotiate industrial-style deals so Australians capture a share of the long-term economic value created by hosting model training, likening the moment to the discovery of natural resources and pointing to sovereign-wealth models as an example. That line of argument has traction within parts of industry and among some former ministers who see a pathway to national benefit if the right commercial terms are struck.

But campaigners and some technologists want stronger rules on data provenance, royalties and public equity in exchange for hosting training infrastructure. Those calls intersect with immediate safety concerns flagged on national television. Four Corners highlighted international litigation and research into harms from chatbots and content-generation systems, including a study that simulated teenage users and found harmful responses from models. The study's lead described how some systems in controlled tests provided detailed instructions for self-harm. An OpenAI spokesperson told the program the company treats the wellbeing of teenagers as a top priority and is strengthening safeguards.

The result is a policy Rubik's cube. Ministers must weigh the short-term gains from investment, including new data-centre jobs and infrastructure spending, against long-term energy and emissions requirements, and the need for clearer legal frameworks to manage harms. The strategy's emphasis on institutions is intended to buy time to build regulatory capability, yet critics say time is the one thing Australia may not have if data-centre buildout outpaces renewable generation and grid upgrades.

Those tensions are now political and practical. Former ministers, civil society groups and some technologists are pushing for stronger rules on data provenance and royalties, while international firms press for stable regulatory and commercial conditions. The Mamre Road site has been flagged by both proponents and opponents as a concrete milestone. Its projected peak emissions in 2032 is the test case campaigners and industry will point to as the buildout proceeds.

Related Articles

The next hard deadline is 2032, when critics say the Mamre Road campus could reach peak emissions. Approval decisions and energy planning for major campuses between now and then will decide whether Australia captures the economic prize of hosting AI training without taking on a large emissions and safety bill.

This article was created with AI assistance.