Australia Tells AI Giants: No Free Training on Our Culture

AI COPYRIGHT

It is rare for a government to say no twice to the same ask. Australia just did. On Monday, Attorney-General Michelle Rowland killed a proposed copyright exemption for AI training for the second time in a year — and this time she did it after her own government's productivity advisers asked for it.

Published October 23, 2026 · By Daniel Rao

Here's the thing about the AI copyright wars: everyone has been waiting to see which country blinks first. The tech giants have spent two years telling every parliament on Earth that their models can't possibly work unless they're allowed to train on everything, for free, forever. Australia, on Monday, became the country that answered back. Twice.

Rowland's announcement — a statement on Sunday night, then a question-time declaration on Monday — ruled out a "text and data mining" exception to the Copyright Act. In plain English: no carve-out that would let AI companies scrape Australian songs, books, journalism and art to train their models without asking or paying. The creative industries called it a landmark. The tech lobby called it, diplomatically, a work in progress. Nobody called it a surprise.

The second "no" in twelve months

This is where the story gets interesting, because Australia has actually been here before. Rowland first ruled out a text-and-data-mining exception in October 2025, telling ABC radio "we will not be entertaining a text and data mining exception". For about a year, that was the end of it.

Then, in August, the Productivity Commission's interim report put the idea back on the table — suggesting a copyright exemption for AI training as part of its recipe for growth, and estimating AI could deliver a $116 billion boost to the economy over a decade. Atlassian co-founder and Tech Council of Australia chair Scott Farquhar championed the proposal in a National Press Club address in July, arguing that "fixing this one thing could unlock billions of dollars of foreign investment into Australia". His logic: under current law, an AI company would have to "cut a deal with every single recording artist in the entire world" to comply, and that complexity was scaring investment away.

The creative sector erupted. Artists, authors and media companies warned the exemption would amount to "industrial-scale copyright theft", as reporting on the backlash put it. And on Monday, Rowland agreed with them — in question time, no less.

"Australian creatives are not only world class, but also the lifeblood of Australian culture, and we must ensure that the right legal protections are in place," she told parliament, in remarks carried by Australian Associated Press. Her written statement was equally blunt: "Artificial intelligence presents significant opportunities for Australia and our economy, however it's important that Australian creatives benefit from these opportunities too. This government has repeatedly said that there are no plans to weaken copyright protections when it comes to AI."

We read the attorney-general's full statement, as republished in full by Australian Arts Review, and checked it against the Copyright and AI Reference Group's published remit. One line stands out as the government's actual instruction to both sides: "The tech industry and the creative sector must now come together and find sensible and workable solutions to support innovation while ensuring creators are compensated."

Translation: no free lunch, but dinner isn't cancelled. Negotiate.

What a text-and-data-mining exception actually is

Strip away the jargon and it's simple. Text and data mining is what AI labs do when they ingest vast piles of material — lyrics, novels, news articles, images — to teach models how language and images work. An "exception" would have made that ingestion lawful by default in Australia: scrape first, ask never, pay nobody.

Australia's current law has no such exception. The Copyright Act 1968 gives rights-holders exclusive rights — reproduction, publication, communication — and infringement arises where an unauthorised use engages those rights unless a statutory exception applies. What Rowland has now done, twice, is refuse to create a new exception for AI. That means AI developers training in Australia need permission — a licence — not just a fast crawler.

It also means something subtler and more important: the default now favours negotiation. A paid collective licensing model — a pool AI companies pay into, with money distributed to rights-holders, much like the music industry's existing royalty machinery — is the shape of the future Rowland's government is sketching.

The two-day meeting that decides what comes next

Rowland convened the Copyright and AI Reference Group (CAIRG) for a two-day meeting this week to thrash out three priority areas, and these are where the actual policy gets built:

First, fair legal avenues for using copyright material in AI — examining whether to establish a new paid collective licensing framework under the Copyright Act for AI, or keep the voluntary licensing status quo.

Second, certainty — clarifying how copyright applies to material generated through the use of AI. If an AI writes the song, who owns it?

Third, cheaper enforcement — making it easier to enforce existing rights through a potential new small-claims forum for lower-value copyright infringement. This one matters enormously to small creators: the right to sue is meaningless if the lawsuit costs more than the stolen song is worth.

One detail worth noting: Annabelle Herd, chief executive of ARIA and PPCA, actually sits on the reference group — so the music industry's voice is in the room, not just in the press releases.

The creatives are cheering — the tech lobby is being diplomatic

The response from Australia's creative and media industries was swift and, frankly, gleeful. Herd called the decision "a landmark decision that reinforces Australia's commitment to its artists and creative industries", adding: "It's not for tech giants to determine how creators' IP is utilised. Protecting that agency is how we build a creative economy that continues to grow, inspire and compete globally." Her colleague at the group called it "a critical first step" toward fair compensation, saying "artists deserve the right to decide how their work is used and to share in the value it creates".

News Corp Australia executive chairman Michael Miller echoed the sentiment, calling the announcement "a welcome catalyst for tech and AI companies to license Australian content". He went further: "By upholding the creator's right to control access, terms of use and payment, it reinforces that our copyright law works to ensure effective market outcomes."

Erin Madeley, chief executive of the Media, Entertainment and Arts Alliance, said in a statement that "the decision backs in the value of our music, screen productions, journalism, books and scripts, and of Australian culture at large".

But the most human moment belonged to Holly Rankin, the songwriter who performs as Jack River. She didn't bother with a press release. She told reporters in Canberra it was an "important decision" — then asked the question that really matters: "What comes next? I believe creators and creatives and Australians should decide where and how their work is used and what they get paid for it." There's something refreshing about an artist saying the quiet part out loud in front of parliament.

On the political side, Liberal senator Sarah Henderson said she was pleased "to see common sense prevail", telling Sky News: "Frankly, there was absolutely no basis for any change in the law which would allow work to be effectively stolen by big tech companies." She also landed a quiet but devastating detail: the Productivity Commission's hearings revealed commissioners had not conducted any economic modelling specifically on data scraping — the $116 billion headline number wasn't built on modelling the actual policy.

The tech side's response was measured. The Tech Council of Australia didn't comment directly on Rowland's move, saying only that it looked forward to working with the reference group "to help develop a framework which we hope will deliver certainty for AI training as well as for artists and creators". Digital industry group DIGI acknowledged the importance of supporting creatives, noted that text-and-data-mining exceptions have been introduced internationally including Japan, and its managing director Sunita Bose said governments worldwide are "exploring how to harness the benefits of responsible AI development and provide appropriate protections to rights holders".

Dean Ormston, chief executive of APRA AMCOS, which represents more than 128,000 songwriters, composers and music publishers across Australasia, was the least diplomatic of the lot: "Tech platforms are shopping for jurisdictions that will water down copyright legislation so that they can find ways to avoid paying creators while profiting from their cultural IP," he said. "It's time for tech companies to stop delaying and start licensing discussions covering both the input and output of creative materials in AI platforms."

A recording studio control room with a large mixing console — music studios are at the centre of the fight over who gets paid when AI models train on songs

Where Australia now sits on the world map

This is the part with genuine global significance, because no country has figured this out cleanly yet. As Marketing-Interactive's reporting on the decision put it, the stance brings Australia "closer to the European Union's more creator-protective approach, and contrasts sharply with the United States, where open use of copyrighted material for AI training remains largely untested in the courts".

The scoreboard, as we can verify it: the EU's 2019 copyright directive created a text-and-data-mining exception but allowed rights-holders to opt out for commercial uses — precisely the model Australia's own broadcaster, the ABC, has argued doesn't work in practice. The UK's approach has been narrower still, with TDM exceptions limited to purposes like research and non-commercial use. Japan went the opposite direction entirely, amending its copyright law in 2018 to let AI companies use copyrighted works for machine learning without restriction — a decision that has drawn heavy criticism from Japan's creative sector ever since. (Both points are well documented in international copyright commentary.)

The United States is the messiest case of all. There is no TDM exception; there is "fair use", argued case by case, at enormous expense. Earlier this month we reported the first US appeals-court ruling touching AI-training fair use — the Third Circuit's rejection of the fair-use defence in the ROSS v Thomson Reuters fight, with a footnote warning the generative-AI industry. Meanwhile the New York Times is fighting OpenAI and Microsoft over what its own internal memos have described as the mass copying of news articles — we reported on those memos last month — and the US Department of Justice has even intervened to back fair use for AI training.

Perhaps the most telling statistic comes from the music industry itself: the International Confederation of Music Publishers says it looked at 62 countries considering law changes on AI training data, and none of the other 61 added a new text-and-data-mining exception for commercial purposes. Australia, it turns out, is not the outlier. It's the consensus — it just said so louder.

Our take: Canberra is setting the price of admission, not pricing itself out

Let's be honest about the scaremongering first. The $116 billion figure is impressive until you notice, as Senator Henderson did, that it wasn't based on modelling data scraping itself. It is a number about AI's general economic potential, wielded as if it were a number about the copyright exception specifically. That is not an argument; it is a fog machine.

The "we'll take our investment elsewhere" threat deserves similar scepticism. Tech companies don't train models where copyright law is friendliest — they train where the compute is. And a model trained anywhere still has to be sold somewhere; jurisdiction over training is not jurisdiction over the market. Australia's leverage is not where the GPUs sit. It's where the songs are.

But — and this is the part Canberra cannot afford to fumble — a refusal is not a regime. "No exception" is a sentence. A paid collective licensing framework that actually moves money from AI companies to Australian songwriters, journalists and authors is a system, and it does not exist yet. The CAIRG meeting this week is where that system either starts to take shape or dissolves into another round of consultation. The Joint Select Committee on AI, which has been hearing testimony from OpenAI and Anthropic executives in recent weeks — including the extraordinary admission that Sam Altman met Australia's deputy prime minister without knowing about the Medicare data breach — delivers its full report on November 30. That's the real deadline.

So here's our position: this decision makes Australia a licensing pioneer, not a market AI labs will quietly skip. The evidence for the pioneering reading is that licensing deals are already proving workable — Google with AAP, OpenAI with The Guardian and News Corp, the major record labels with AI platforms — and nobody has shown that a free-scraping exception was ever the binding constraint on AI investment. But the pioneer label only sticks if Canberra builds the licensing machinery it promised. A principled "no" with no plumbing is just a speech.

Update, Oct. 23: Australia's 'no' just went global — 63 media groups join the opt-out fight

Canberra's "no" has left the building — and it is picking up company. Media Today and The Will reported on October 21 that an international coalition of 63 media groups is now fighting the AI copyright opt-out together, and their argument is a direct challenge to the way the training era has worked so far: "A creator should not have to discover that an AI company wants their work and then take action to stop it. The AI company should have to ask first." Opt-in, in other words, is becoming a global demand — not an Australian quirk.

The front is widening in both directions at once. Nigeria is now probing Big Tech over how it uses local content — a development the Australian decision helped legitimise, reading Canberra's refusal as the first domino rather than a lone outlier. And Marketing-Interactive's fresh relay of the Rowland ruling — the attorney-general's statement, the three CAIRG workstreams of paid collective licensing, AI-generated-content copyright and small-claims enforcement — confirms the story is still moving through international coverage rather than settling.

The usual caveat still applies: this is a fight over what the rules could become, not a finished copyright regime. The opt-in versus opt-out distinction is the live policy war, and the coalition is demanding opt-in by default — the strongest possible version of the position Australia just took. As this post argued, Australia is not the outlier. It is the consensus — and the consensus is growing.

FAQ

What is a "text and data mining" exception?

It's a legal carve-out that would let companies copy copyrighted works — articles, books, images, music — for automated analysis without the rights-holder's permission. For AI developers, it would have meant training models on Australian creative works for free and by default. Australia has now refused to create one, twice.

Hasn't Australia already said no to this?

Yes — Attorney-General Michelle Rowland first ruled it out in October 2025. The Productivity Commission revived the idea in an August interim report this year, which is why Rowland had to kill it again, in question time, on October 19, 2026. Same answer, louder delivery.

Does this make AI training illegal in Australia?

No. It means there is no automatic legal shortcut. AI developers must still operate within existing copyright law — which generally requires permission (a licence) to reproduce copyrighted works. The government is now working, through the Copyright and AI Reference Group, on licensing frameworks to make that permission processable at scale.

What happens at the two-day CAIRG meeting?

The Copyright and AI Reference Group is examining three things: whether to build a new paid collective licensing framework for AI under the Copyright Act; how copyright applies to AI-generated content; and how to make enforcement cheaper for small creators, potentially via a new small-claims forum.

How does Australia's stance compare to the US and EU?

The EU has a TDM exception with an opt-out for commercial uses; the US has no TDM exception but is litigating "fair use" for AI training case by case (including the ROSS and New York Times cases). Australia's position — no exception, permission required — is stricter than both, and closer to the creator-protective end of the spectrum.

Who wins and who loses?

Short term: Australian creators, publishers and rights-holders, who keep control and negotiating leverage. Short term losers: AI labs that wanted legal certainty for free training data in Australia. The real test is whether the promised licensing framework arrives — if it doesn't, both sides lose to litigation instead.

Sources: We relied on reporting from Marketing-Interactive, Channel News Australia, Australian Associated Press (via Feedsy News), Australian Arts Review (which republished the attorney-general's statement in full), Music Business Worldwide, Tech Times, the Hugh Stephens copyright blog, Spruson & Ferguson's IP analysis, and Mubert's briefing on the ICMP position. Quotes are attributed to the outlet that carried them. Internal links are to our own earlier reporting, verified live this week.

Search This Blog

TechPulse USA