Annabelle Herd, CEO, ARIA
If you work in Australian media and you haven’t been paying close attention to the copyright and AI debate beyond the News Bargaining Incentive, it’s time to start: the outcome of this fight will determine whether the content your organisation creates retains its commercial value or gets strip-mined by trillion-dollar systems for free.
Australia’s media and creative landscape had a huge win in October, when Attorney-General Michelle Rowland ruled out a text-and-data-mining exception for AI training in a global first.
You’d be forgiven for thinking that settled the matter.
While it was a huge step forward, it was the beginning of a much longer fight to defend Australian rights holders’ entitlement to consent and remuneration for IP. If anything, the lobbying has intensified since then, and it’s gotten slipperier.
The direct pitch for a copyright exception has been replaced by softer language about “investment signals” and “regulatory settings” but the destination is the same.
Most recently, this came to the fore at the India AI Impact Summit last month, where over AU$282,000 in AI investment commitments were announced, and a familiar chorus resumed here at home.
Technology lobbyists and sympathetic commentators are again warning that Australia made a mistake by rejecting a text-and-data mining exception to our copyright laws. The argument: without preventing AI companies from scraping creative works without permission, investment will bypass us. It sounds urgent, maybe even plausible, but it falls apart the moment you look at what AI companies are doing in the real world.
Australia is not the outlier we’re made out to be: the United States has no text-and-data mining exception and currently has over 60 AI training lawsuits on foot.
Canada and New Zealand have no such exceptions either. The UK proposed one in 2024, which was met with fierce opposition from its creative sector, and a recent parliamentary report recommended following Australia’s lead.
Where exceptions have been introduced – the EU in 2019, Japan and Singapore before ChatGPT – none saw a resulting surge in investment. The capital has gone where licensing markets operate, because licensed content delivers what AI companies actually need: defensibility.
Licensing deals are being signed at a pace across every content vertical, around the world. They exist because copyright law gives rights-holders something to negotiate with, and they prove that existing copyright legislation is more than capable of working with AI the same way it works with every other industry.
News Corp signed a multi-year global partnership with OpenAI covering Australian mastheads including The Australian, The Herald Sun and The Daily Telegraph. AAP struck a licensing deal with Google for AI search and summarisation. HarperCollins entered into agreements covering Australian-published authors. Getty Images partnered with Canva on a responsible AI training framework that encompasses Australian imagery.
In music, all three global majors – Universal Music Group, Sony Music and Warner Music Group – have signed licensing deals with AI platforms, including Klay and Suno, whose catalogues include Australian repertoire.
Merlin, the independent network that includes Mushroom Music, signed with Udio. Kobalt and Merlin signed with ElevenLabs, so artists and songwriters can participate in model training and receive royalties. Adobe signed with talent agencies CAA, UTA and WME to supply training data for its Firefly Foundry AI, these are agencies representing Australian creators.
Why does that list matter? Because every one of those deals is proof that copyright and AI development work together.
Licensed content provides AI companies with legal certainty, high-quality training, and defensible products. It gives creators income, control and agency over how their work is used. A licensing market functioning at this scale tells you something important: the most significant wave of technological innovation in a generation is being built on the foundation of intellectual property rights.
The system works.
Arguments designed to save hyperscalers a quick buck under the thin veil of innovation cannot be allowed to tear that system down at the expense of our media and creative industries.
Companies are spending AU$1.8 trillion globally on AI infrastructure this year. Estimated global content licensing costs sit at AU$4.3–A$5.7 billion. Licensing is a rounding error on these companies’ balance sheets. Framing it as a barrier to progress is a bid to shave costs at the margins of an industry awash with capital, and it would come at the direct expense of Australia’s $67 billion creative economy and the roughly $900 million our collecting societies distribute to creators every year.
AI companies argue that copyright protections should be relaxed so their systems can train more freely and cheaply.
At the same time, those same companies tightly control how their own AI systems are deployed: strict contracts, enforceable protections, the lot. They want frictionless access to the content that powers their products, and ironclad control over the products themselves. We’re asking for the same principle in return. Permission, control and enforceable terms.
If you’re a media executive, an editor, a journalist – anyone whose livelihood depends on the commercial value of content – this debate has direct consequences for your business.
The 2025 decision to reject a text-and-data mining exception gave Australian media and creative industries a foundation to negotiate from. Any other access scheme proposed by the government would undermine our right to negotiate terms of use and devalue the investment in our intellectual property. We back ourselves every time to negotiate our own deals better than the government can.
We cannot afford to reopen this question, undermine deals already in progress and tell the world that Australian content is there for the taking.
Copyright works.
Licensing works.
The deals prove it.
Now we need to make sure the framework that enabled them stays in place.