Australia’s AI Copyright Fight Is the Deal Every Democracy Will Be Asked to Sign

Opinion | Technology and civic life

Australia is heading into one of the cleanest tests yet of the AI era: whether a democratic government can welcome the data-center boom without quietly treating writers, artists, musicians, journalists and other creators as the raw material that makes the boom cheap.

The immediate fight is in Canberra. Prime Minister Anthony Albanese is expected to give a major speech on artificial intelligence this week, according to Guardian Australia, as his government tries to balance two promises that are starting to collide. One promise is economic: Australia wants a place in the global AI infrastructure race, including the data centers, energy contracts and foreign investment that now follow frontier model companies around the world. The other promise is civic and cultural: Australians who make books, songs, journalism, screen work and visual art should not lose control of their work just because a powerful software company can scrape it faster than lawmakers can write rules.

That is not a niche copyright argument. It is a governance argument wearing a copyright hoodie.

Guardian Australia reported Saturday that authors, artists, musicians and media organizations fear the government could revive a “text and data mining” carve-out that would let AI companies scrape content to train models without infringing Australian copyright law. The government says it has no plans to grant such an exemption. Albanese, asked last week whether copyright safeguards were at risk, said these are “complex issues” and pointed to his government’s “strong record” of supporting people who create things, including the principle that if work is used, creators should be “properly compensated.”

That reassurance matters. It also needs to be converted into policy before the lobbying pressure gets louder.

The report lands at an awkward and revealing moment. AI firms and their allies are arguing, in Australia and elsewhere, that legal certainty around training data is a condition for investment. In the Australian case, the Tech Council of Australia chair and Atlassian co-founder Scott Farquhar previously argued that “fixing this one thing” could unlock billions of dollars in foreign investment, Guardian Australia reported. The Productivity Commission floated the idea of a text-and-data-mining exemption last year, triggering backlash from the creative sector before Attorney General Michelle Rowland killed off the proposal in October and began fresh consultations on copyright modernization, including a paid licensing model.

Now the fear is that the same idea could return in softer packaging. Independent Senator David Pocock has said his office was tipped off about an industry push for a copyright carve-out tied to at least $50 billion in data-center investment and a proposed creative fund worth $350 million a year. The government rejected his claims as inaccurate and repeated that it had no plans to weaken copyright law. Guardian Australia also reported that the Australian Financial Review said Anthropic was pushing for a deal in line with what Pocock alleged, while industry and government sources played down the existence of such a deal.

The details are still contested. The underlying pressure is not.

Every country that wants AI investment is being handed a shiny version of the same bargain: give us cheap energy, land, favorable regulation and legal room to train; we will give you infrastructure, jobs, innovation branding and a seat near the future. Some of that bargain is real. Data centers are physical, expensive and strategically important. AI infrastructure will shape productivity, defense, media, education, health care and software markets. A country that says no to everything will not become more sovereign; it may just become more dependent on systems built somewhere else.

But a country that says yes to everything is not doing industrial policy. It is doing procurement for someone else’s business model.

Australia’s debate is especially useful because it refuses the fake binary that dominates AI politics. This is not simply “artists versus innovation.” It is not “copyright absolutists versus the future.” It is not even “local culture versus American tech,” though the national-sovereignty layer is hard to miss. The real question is whether the public should accept a rule in which creative labor is privately monetized after being publicly devalued.

AI companies often talk about training data as if it were atmospheric: everywhere, ownerless, naturally available, inconvenient only because the law has not caught up. That framing is doing a lot of work. Books are not air. Journalism is not weather. Music is not a mineral deposit. The fact that creative work is easy to copy does not mean it is socially free. It means society has to decide whether ease of copying becomes a license to ignore the people who made the work valuable in the first place.

That decision cannot be outsourced to venture-backed urgency.

The strongest argument for a text-and-data-mining exception is not ridiculous. AI developers need large, diverse datasets. A licensing thicket could entrench the biggest incumbents, because only the richest companies can negotiate with everyone. Public-interest researchers, startups and local companies could be locked out while the global giants that already scraped the open web move on. If every use of publicly available text becomes a slow, expensive rights negotiation, innovation could narrow rather than broaden.

That is the steelman. It is also why “just enforce copyright exactly as before” is not a complete policy answer.

But the policy answer cannot be “therefore, creators eat the cost.” If the law creates an exception, it is choosing winners. If it refuses collective licensing or compensation mechanisms, it is choosing winners. If it treats training as magically different from every other value-extracting use of work, it is choosing winners. The only honest question is who gets protected, who gets paid and who gets told that their bargaining power vanished in the name of progress.

Australia’s government appears to understand at least part of this. Guardian Australia reported that its stated preference is for tech firms to negotiate agreements with creatives to pay for use of their content. That is the right starting point. But a preference is not a structure. “Go negotiate” works poorly when one side is a novelist, an illustrator, a local newsroom or a small music label and the other side is a frontier AI company backed by billions of dollars and an implicit threat to build somewhere else.

If Albanese wants a serious AI agenda, the answer should be neither a blanket carve-out nor a vague promise that creators will somehow get paid later. Australia should set terms now: no uncompensated mass scraping exception for commercial foundation-model training; a transparent licensing pathway for rights holders who want to participate; collective mechanisms so small creators are not forced into impossible one-by-one negotiations; disclosure duties that let the public know what categories of work were used; and separate rules for public-interest research, archives and genuinely noncommercial uses.

That kind of framework would not be anti-AI. It would be pro-market in the grown-up sense: markets need property rights, bargaining rules and enforceable duties. “Move fast and ingest the culture” is not a market principle. It is a power move.

The data-center half of the story makes the stakes even clearer. Australia is attractive to AI infrastructure companies because it is politically stable, has land, and has access to renewable power. But data centers are not frictionless gifts. They require electricity, water planning, transmission capacity, local approvals and, increasingly, public patience. Former industry minister Ed Husic has warned that Australia should remember it has negotiating leverage and set terms rather than make an impulse purchase. Assistant Minister for the Digital Economy Andrew Charlton has said Australia should not “blindly accept or reject” investment but should set terms aligned with national values and long-term interests.

That is exactly the right frame. The only problem is that “values” get very slippery when a large investment number enters the room.

A data-center investment package tied, directly or indirectly, to weaker copyright protection would send a brutal message: we value land, energy and compute enough to negotiate hard, but we value creative work enough to put it in the sweetener pile. That would be bad politics and worse economics. Creative industries are not decorative. They are part of the information supply chain that AI systems depend on. Journalism, books, scripts, music, images, code documentation, research summaries and cultural criticism all make models more useful. If the AI economy depends on those inputs, then paying for those inputs is not charity. It is cost recognition.

There is also a trust problem. Voters are already uneasy. Guardian Australia cited a May Guardian Essential poll finding that 36% of voters see AI as carrying more risk than opportunity, 41% see risk and opportunity as about the same, and only 22% see more opportunity than risk. A government that tells a skeptical public that AI will be managed in the national interest cannot then appear to trade away creators’ rights behind closed doors. That is how useful technology becomes politically toxic.

The better path is boring and durable: make the bargain public.

If AI firms want access to Australian creative work, they should say what access they want, what they are willing to pay, what transparency they can provide, and how Australian creators can opt in, opt out or challenge misuse. If the government wants data centers, it should say what energy, water, land-use, labor and community-benefit standards apply. If officials believe copyright law needs modernization, they should explain which uses are being protected, which are being compensated and why. No mystery boxes. No “trust us” carve-outs. No innovation theater.

The temptation for governments is to treat AI policy as a race in which hesitation equals defeat. That is too simple. Speed matters, but so does posture. Countries that set clear rules early may be more attractive than countries that offer uncertainty plus political backlash. Companies building long-term infrastructure do not only need permissive rules; they need stable ones. A rushed copyright giveaway could trigger years of litigation, elections fought over tech capture, and a creative-sector revolt that makes future AI policy harder.

This is why Australia’s fight should be watched well beyond Australia. The U.S., Europe, the U.K., Japan and smaller democracies are all circling versions of the same question. Who pays for the human-made material that made generative AI impressive? Courts will answer part of it. Markets will answer part of it. But governments cannot duck the distributional choice.

My view: Australia should welcome AI investment and reject the shakedown logic. Build the data centers if they meet energy, water, housing and community standards. Court the frontier labs if they respect local law. Create licensing systems that are practical enough for startups and strong enough for creators. Fund public-interest AI research. Modernize copyright where it genuinely blocks socially useful work. But do not legalize mass uncompensated extraction and call it a national strategy.

The future does not become more innovative when the people who make culture are told their work counts only after a model has learned from it.

Albanese’s speech this week may not include a concrete copyright announcement. That is fine if the government uses the time to build a better rulebook. It is not fine if delay becomes a holding pattern while industry tries to rebrand a carve-out as competitiveness.

The clean principle is simple: AI companies can build in democracies without being allowed to quietly rewrite the social contract. If they need the culture, they can pay for the culture. If they want public infrastructure, they can accept public terms. And if governments want to be taken seriously as stewards of the AI transition, they should stop treating creators’ rights as the negotiable line item in someone else’s data-center pitch.

Australia has leverage. The question is whether it will use it.

Sources and attribution

• Guardian Australia reported on July 12, 2026, that Albanese is expected to deliver a major AI speech this week, that Labor ministers are split over copyright reform, and that creators fear a revived text-and-data-mining carve-out tied to the data-center investment race.
• Guardian Australia quoted Albanese saying his government supports creators having control over what they create and being properly compensated when their work is used.
• Guardian Australia reported Pocock’s allegation of an industry proposal involving at least $50 billion in data-center investment and a $350 million-a-year creative fund, and also reported the government’s rejection of that characterization.
• Guardian Australia cited a May Guardian Essential poll finding 36% of voters see AI as more risk than opportunity, 41% see risk and opportunity as about equal, and 22% see more opportunity than risk.