For late 90s and 2000s kids who were already online, one of the biggest scare campaigns was the forbidden use of pirating software like Napster or Limewire. There was a persistent threat that if you downloaded the wrong song, a US record label would emerge from the dial-up ether and bankrupt your entire bloodline. Not only were big names like Metallica suing these platforms, but the recording industry was actively litigating against individual downloaders. Related Article Block Placeholder Article ID: 328096 Neural Notes: Why Germany’s landmark AI copyright case is a warning for Australian companies Tegan Jones There were persistent news reports about teens getting hit with eye-watering damages over a few dozen MP3s. Anti-piracy ads treating a copied song like grand theft auto. Elder millennial Australians, in particular, still have the phrase ‘you wouldn’t steal a car’ burned into our collective consciousness. The message was simple and deeply moralistic: copying is theft, and the internet is a crime scene unless tightly policed by the people who own the content. Smarter business news. Straight to your inbox. For startup founders, small businesses and leaders. Build sharper instincts and better strategy by learning from Australia’s smartest business minds. Sign up for free. * indicates required Email Address * By continuing, you agree to our Terms & Conditions and Privacy Policy. Most of us never ended up getting sued, of course. But that was never really the point. The fear worked because it was ambient. It lived in the culture. Copyright enforcement was personal and theatrical. Which is why the current fight over AI copyright in Australia feels so surreal by comparison. From kids with MP3s to killing Pirate Bay A few years later, the panic was no longer just about teenagers with dodgy MP3 folders. It was about the plumbing of the internet itself.  The Pirate Bay arrived on the scene almost like a dare to international authorities. It was an index for just about everything, wrapped in a skull-and-crossbones brand and a gleeful sense of online punk. It wasn’t just that you could find almost any movie, album, or game. It was the swagger that emanated from those who ran it. The response was harsher than the Napster era in some ways, and more structural in others.  Related Article Block Placeholder Article ID: 269876 Generative AI is the next chapter of copyright problems for authors Tegan Jones The Pirate Bay founders were prosecuted in Sweden, convicted of assisting copyright infringement, sentenced to prison, and ordered to pay substantial damages.  In Australia, rights-holders tried to make ISPs responsible for BitTorrent piracy through a very public iiNet case. They lost, incidentally, before pivoting to a more direct strategy: site blocking.  By the end of 2016, Australian courts had ordered ISPs to block The Pirate Bay and other torrent sites, forcing providers to “take reasonable steps” to disable access. For users, it became a bit of a game to see where The Pirate Bay would pop up next. There were new domains, new proxies, new workarounds.  And in the end, The Pirate Bay never really died. It was blocked at the ISP level in Australia, but something as simple as a VPN has kept it accessible for anyone determined enough.  This has been important because it continues to prove that two things can be true at once: governments were willing to reroute the internet in the name of copyright protection, and the internet remained stubbornly hard to police. If you stopped the story there, it would all seem fairly coherent. First, the scare campaigns. Then the platform lawsuits. Then the blocking orders.  Taking creative work without paying for it is stealing, and copyright law is the weapon used to bring the internet – and sometimes even individuals – to heel. The moment when copying work became ‘infrastructure’ Somewhere between the court orders to block The Pirate Bay and ChatGPT landing on the scene, the politics of copyright changed.  It very quickly became about scale and subsequently the way policymakers talk about it. Entire AI systems were trained on mountains of text, images, code, and audio pulled from across the internet, including material whose owners had spent years being told the law would protect them.  And instead of governments talking about blocks, shutdowns or criminal sanctions, they started talking about licences. By the time most people touched an LLM, the copying horse had well and truly bolted. OpenAI, Meta, Google and others had already ingested huge quantities of books, news, reference works and other copyrighted material, including Australian content, of course. Lawsuits did follow, and they’re still piling up. But they arrived long after the models had been built, launched and embedded into products — and paid for by businesses and individuals. Related Article Block Placeholder Article ID: 141869 Australian tech startups stand to lose out in proposed copyright reforms The Conversation And that’s what distinguishes the modern copyright fights from the Napster and Pirate Bay eras. Back then, the law was deployed to, theoretically, stop the behaviour. But now it’s being used to work out what to do after it’s already happened and been deemed largely socially acceptable. And that’s not to say the allegations are trivial. Major AI companies have been accused of training on the likes of pirated books, shadow libraries, archives and news sites without permission.  Now courts and policy bodies have to wrestle with whether AI training on copyrighted works is fair use (which OpenAI and Google have openly lobbied for), whether pirated sources matter, and when outputs cross the line into infringement.  But the unspoken truth is that nobody — even those vehemently vocal against this practice — realistically expects these LLMs and chatbots to be deleted or meaningfully unwound at this point. So what we’re left with are ‘practical’ remedies such as disclosures and settlements. And perhaps in some cases, settlement payouts, but not nearly enough of them. Canberra’s AI copyright plan For its part, Australia has quietly been sketching out what that post-facto settlement might look like.  The Albanese government has publicly ruled out a broad text-and-data-mining exemption for AI training, insisting it will not simply let developers use Australian creative work for free.  Instead, the official direction of travel has been toward licensing under Australian copyright law. This would preserve the idea of permission and payment, but adapt it for an AI market in which nobody is realistically going to negotiate one-by-one with every writer, artist, publisher or musician whose work sits inside the training machine. The Copyright Agency has already started building the rails for that world. It extended its Annual Business Licence so licensed workplaces can use news media content in prompts for AI tools, and in 2026 began consultations on licensing member content for AI activities, including training, fine-tuning, and retrieval-augmented generation.  That is a remarkable shift in tone if you are old enough to remember Limewire. What would once have been framed as piracy is increasingly being reframed as infrastructure that can be sold back to people, provided the right collective licence sits underneath it. In other words, we’re going wholesale. Related Article Block Placeholder Article ID: 285543 Canberra steps into the AI arena without a policy framework Jayson Lamchek Pocock, Ayres and the alleged secret AI copyright deal But just when this debate seemed to be leading to something technocratic and opaque, a figurative Molotov cocktail was thrown in Parliament this week. Independent senator David Pocock alleged a whistleblower had told him the federal cabinet was considering two competing options for AI and copyright. One is a copyright carve-out for AI firms tied to billions in data-centre investment and a large creative fund. The other is an expansion of licensing arrangements that would clear the way for AI companies to train on Australian material as long as they pay.  Industry minister Tim Ayres denounced the claims as “reckless speculation,” and government spokespeople repeated the line that Labor had already ruled out a text-and-data-mining exception. What’s interesting here is that if these choices have even a whiff of truth, it means the choice isn’t really between copyright and not to copyright. It’s between two different ways of normalising mass AI training on Australian content. Either a carve-out dressed up as economic development, or a licensing system dressed up as consent. It’s worth noting that Ayres did make a broader strategic argument quite clear.  The minister said Australia can’t afford to be “a cork bobbing on the ocean of other people’s technology” and should secure as much of the technology stack as possible.  It’s not a new perspective – we’ve been hearing about the need for national AI capability, investment and sovereignty for quite a while now. But it does reveal a deeper compromise that is underway. Once the AI stack is treated as inevitable, the real policy fight is no longer about whether the copying should have happened. It is about what Australia, as a nation, can extract in return. What Australia’s AI copyright plans mean for small businesses Small businesses are now caught between both ends of this system. Founders, publishers, agencies and solo operators are part of the pool of people whose work may already have been swept into training datasets without any real say in the matter.  But they’re also under pressure to use AI tools themselves to move faster, cut costs and stay competitive. And when they do, Australian copyright law offers them a pretty awkward deal.  Related Article Block Placeholder Article ID: 255488 If you’re using AI to create business assets, they’re probably not covered by Australian copyright law Tegan Jones Legal guidance consistently notes that copyright protection here depends on human authorship and independent intellectual effort. Purely AI-generated outputs may not attract copyright at all, because the “author” is not legally human. That means an SME using generative AI for any kind of creative output – blog posts, ad copy, design concepts, product descriptions, etc – may be relying on material they don’t fully control or clearly own.  At the same time, those same businesses are warned their use of AI can create infringement risk, whether by uploading protected material into tools, training on scraped material, or publishing outputs that reproduce a substantial part of existing works.  It’s created a strange asymmetry that has simply been accepted. Small businesses can become involuntary inputs into the AI economy, then become paying customers of that economy, and still end up with shakier rights over what comes back out. This is why the current debate over AI copyright in Australia matters beyond artists, publishers or Silicon Valley giants.  For Australian startups and SMEs, the argument over carve-outs versus licensing is about whether the next layer of digital infrastructure will be built on terms they had no role in setting, and whether they will bear the compliance costs of a system designed around much larger players.  A bespoke licence might reduce legal uncertainty for a multinational using AI inside the workplace. But it doesn’t really answer the underlying question of fairness for the small publisher whose archives helped train the model, or the startup founder who discovers the AI-assisted materials their team made are not defensible as proprietary IP. Too big to delete, too late to ask permission And that brings us full circle. In the Napster years, we were told copyright had to be enforced ruthlessly because otherwise culture would collapse.  In the Pirate Bay era, the Australian government was willing to block sites and reshape internet access to prove its dedication to copyright protection.  In the AI era, the biggest copying event in digital history has largely been treated as something to regulate after the fact, whether through collective licensing, creative-industry funds, or investment trade-offs. For those of us who witnessed all of this, it’s flabbergasting.  We went from “copying is theft” to “copying is negotiable”, provided the company is big enough, and the government can spy an economic upside. Stay in the know Never miss a story: sign up to SmartCompany’s free daily newsletter and find our best stories on LinkedIn.