'Not Up for Grabs': Albanese Draws the Line on AI Copyright — But Can Australia Enforce It?

Australian Prime Minister Anthony Albanese has established a new Office of AI within his department and delivered the strongest copyright protection pledge from any world leader on AI, declaring that “not everything produced in Australia is up for grabs” and that no company should use Australian books, music, art, or news to train AI without the creator’s control. The speech at the University of Sydney on July 15 was the culmination of weeks of escalating pressure from Australian creatives — and the first concrete government response to the $50 billion copyright carve-out proposal we reported earlier this month.

🔍 THE BOTTOM LINE

Albanese’s speech was rhetorically the strongest position any government has taken on AI copyright: no text-and-data-mining exemption, no free use of Australian creative work for AI training, and explicit endorsement of creator ownership and compensation. But the enforcement mechanism is legislation that will not reach parliament until early 2027, against companies that have already demonstrated they can set their own terms and prices for national governments. The gap between the rhetoric and the enforcement timeline is where the real fight will happen.

What Albanese Actually Said

The speech contained three major commitments. First, copyright: “Australian writers, musicians, artists and journalists must retain ownership and control of their work. Our laws will spell that out, plain as day.” Albanese rejected the prospect of large companies like OpenAI and Anthropic being given free use of Australian data, framing unauthorised AI training as theft. “No company should use Australian books, music, art or news to build or train AI without the artist’s control. That includes the artist’s control of the price and value of their work.”

Second, datacentres: the federal government will work with state premiers on new planning controls ensuring datacentres do not compete with housing for land, do not dominate local energy systems, and that operators pay for new water infrastructure needed for cooling. Surging datacentre demand will not be allowed to push up household electricity bills. A meeting of national cabinet next month will address the planning framework.

Third, institutional structure: a new Office of AI within the Department of the Prime Minister and Cabinet, with legislation expected in early 2027. Albanese said Australia must be more than “a data warehouse for AI products made overseas” and should benefit from its own innovations.

The Copyright Fight We’ve Been Covering

This speech is the direct follow-up to the story we have been tracking since early July. On July 1, Australian creatives raised the alarm over a leaked industry proposal that would trade Australian copyright protections for $50 billion in datacentre investment — a deal artists called “the ultimate dirty deal.” Powderfinger, Spiderbait, Middle Kids, and other major Australian musicians publicly demanded Albanese block the carve-out.

On July 12, we reported that the copyright battle had reached boiling point, with Labor internally divided and Anthropic actively lobbying for a text-and-data-mining exemption. Albanese’s speech on July 15 was the answer to that pressure — and it was unambiguous on the principle, even as the legislative mechanics remain vague.

Labor has long ruled out a text-and-data-mining exemption allowing AI firms to train on Australian content without compensation. But there has been intensive lobbying from big tech, and cabinet discussions on copyright reform are continuing with a diversity of views among senior ministers. The speech was the strongest assurance yet — but “assurance” is not “law.”

Can Australia Actually Enforce This?

The Guardian’s analysis piece on the speech was blunt about the enforcement challenge. The challenges of regulating social media or stopping hate speech — both areas where the Australian government has struggled — show that these firms “can set their own terms and prices for countries like Australia.” If a government cannot reliably compel a social media company to remove harmful content, compelling an AI company to pay for training data it has already ingested is a harder problem by orders of magnitude.

The structural issue is jurisdictional. AI models are trained on infrastructure in the United States, using data scraped globally. Once a model has ingested Australian creative work, the damage is done — you cannot un-train a model. The only enforcement lever is preventing future training, which requires either API-level auditing (difficult, proprietary), data provenance tracking (does not exist at scale), or trade-related pressure (untested against companies with market caps larger than Australia’s GDP).

Google’s ongoing publisher lawsuit illustrates the pattern: the legal system can punish after the fact, but it cannot prevent ingestion in real time. And the companies know this.

The Datacentre Angle

The datacentre rules may prove more enforceable than the copyright rules, because datacentres are physical infrastructure subject to planning law. Australia can require datacentre operators to pay for water infrastructure, locate away from housing, and absorb their own energy costs — these are local permits, not global policy. The $50 billion datacentre investment pledge that tech companies dangled as leverage for the copyright carve-out may still arrive, but on Australia’s terms rather than as a quid pro quo for weaker IP protections.

NZ Angle

New Zealand faces the same copyright and datacentre questions as Australia, with less leverage. Australia can at least offer a market large enough to justify datacentre investment — NZ cannot. The NZ Super Fund’s sovereign AI infrastructure ambitions and the broader sovereign AI build strategy are NZ’s equivalent of the “data warehouse for AI products made overseas” risk that Albanese named. If Australia succeeds in requiring AI companies to pay for training data, NZ creators benefit through precedent — but if Australia fails, the lesson for smaller economies is that copyright protection against AI ingestion is a principle without a mechanism.

❓ FAQ

Did Albanese ban AI training on Australian content? Not yet. The speech was a policy commitment, not a law. Legislation is expected in early 2027. Until then, AI companies can legally train on Australian content under existing fair dealing provisions, though Labor has ruled out creating a new text-and-data-mining exemption.

What is the Office of AI? A new unit within the Department of the Prime Minister and Cabinet that will coordinate AI policy across federal agencies, work with state governments on datacentre planning, and develop the legislative framework for parliament’s consideration in early 2027.

How does this affect NZ? NZ has no equivalent AI copyright legislation in progress. If Australia’s model succeeds, it creates a regional precedent NZ could adopt. If it fails or is watered down, the lesson is that even mid-sized economies struggle to enforce IP rights against AI companies — which is worse news for NZ’s smaller market.

What about the $50 billion datacentre deal? Albanese’s speech did not reject the investment — it rejected the condition. The datacentre investment can proceed under the new planning and environmental rules. What Australia will not do is trade copyright protections for infrastructure spending.

🔍 THE BOTTOM LINE

Albanese drew the right line: no copyright exemption for AI companies, no free use of Australian creative work, strict datacentre rules. The line is drawn in rhetoric now and in legislation in 2027. Between now and then, every Australian book, song, article, and artwork scraped into a training corpus is gone — and no speech, however strong, will get it back. The real test is not whether Australia can pass the law. It is whether the law has teeth when the defendant is a company with more revenue than the country pursuing it.