iiNet vs. Hollywood: The copyright victory protecting Australian creators against AI
When iiNet defeated Hollywood studios in the High Court, the ruling turned on a crucial distinction: providing internet access was not the same as providing infringing content or the means of infringement. But generative AI may not fit so neatly on the same side of that line.

In 2012, Perth-based internet service provider iiNet walked into the High Court of Australia to take on Hollywood's biggest film and TV studios, and won a case that would redefine Australian copyright law for the internet age. The case, Roadshow Films Pty Ltd v iiNet Limited [2012] HCA 16, found that iiNet was not responsible for its customers using the internet to infringe on the copyright of Hollywood studios.
Fourteen years later, the question of copyright ownership and infringement is at the centre of AI platform development.
This article is the first in a series examining the history of Australian copyright law, how our world-leading copyright protections have been amended in ways that benefit AI companies, and what that means for Australian creators and the majority of 20th century works.
A case that defined Australian copyright for a generation
In the early 2000s, before streaming and long before generative AI, peer-to-peer file sharing was ubiquitous. Millions of people were downloading music and films without paying for them. As studios watched their revenues erode, they went looking for someone to hold accountable. Their target was internet service providers.
In particular, the Australian Federation Against Copyright Theft (AFACT) — acting on behalf of 34 of the biggest film and TV studios including Village Roadshow, Universal, Warner Bros., Paramount, and Seven Network — decided to sue Australia’s third-largest internet provider, iiNet.
AFACT’s argument was that iiNet had “authorised” the copyright infringement of its users. Those users were distributing and downloading films via BitTorrent, a peer-to-peer file-sharing software.
The federation’s argument rested on a well-established principle in Australian copyright law: a party can be held liable for an infringement it did not personally commit, if it had the power to prevent it and failed to take reasonable steps to do so.
AFACT had sent iiNet numerous notices identifying customers it believed were using BitTorrent to distribute films illegally, and demanded the company warn, suspend, or disconnect them.
iiNet did not comply with the demands, and AFACT responded by taking the company to court.

High Court of Australia in 2012. Source: Alex Proimos via Wikimedia Commons
A unanimous victory
The legal battle lasted four years as it made its way through Federal Court, Full Federal Court, and the High Court. But when the judgment came on 20 April 2012, it was unanimous.
The High Court ruled in iiNet's favour. The reasoning came down to iiNet not providing the films, nor the software used to copy them.
Supplying internet access, the Court held, was not the same thing as supplying the means of infringement. The ISP had no direct relationship with its customers’ infringing acts, no control over BitTorrent itself, and no technical mechanism to selectively block infringing traffic without also disrupting legitimate use.
Michael Malone, iiNet’s founder who led the company throughout the legal battle and appeared as an expert witness, said iiNet spent $8.9 million in direct legal fees and recovered a fraction of that.
“More importantly, it was an enormous distraction for four years as it worked its way through the primary case and then two appeals," he told The Discloser.
“Many of our investors saw it as potentially existential. If our legal fees were that high, what might we have had to pay the other side if we had lost, plus any penalties and future compliance costs?”
Michael Malone, founder and former CEO of iiNet. Image: Supplied.
In Malone’s view, the fight was also a matter of principle, not just survival. He still regards the case as one of the company's most important contributions to the internet age.
“The law on copyright liability hadn't been tested in court for the internet age,” he said.
“We successfully argued that iiNet didn't provide the content, movies on the internet, nor the means of infringement, the BitTorrent software.
“Access to the internet could not reasonably be the means of infringement.”
An internet provider, he argued, was not in the business of policing the internet. It was in the business of providing access to it. To hold an ISP liable for the acts of its users would misunderstand the role of a network intermediary, and set a precedent that could expose any provider of general-purpose infrastructure to unlimited copyright liability.
The answer to piracy was not enforcement through ISPs, but by making it easy and affordable to access copyrighted content legally, according to Malone.
“Today, with music and video streaming services being so mainstream, I reckon that [argument] has been mostly proven true,” he said.
Before streaming, iiNet’s “freezone” service offered users unmetered access to licensed material from the ABC, Apple's iTunes, FetchTV and the English Premier League. It was an attempt to build a legal pathway for digital content consumption at a time when much of the industry was insisting the only answer to piracy was enforcement.
Providing infrastructure vs. carrying out infringement
The ruling crystallised a distinction that would reemerge through subsequent technology law debates: there is a difference between providing infrastructure through which infringement becomes possible and providing the means by which infringement is actually carried out.
The most relevant case law at the time of iiNet’s legal battle was Moorhouse v University of New South Wales [1975] HCA 26. The High Court found UNSW liable for copyright infringement because it had supplied both the content (the books) and the means of copying them (the photocopiers). The combination of content and copying mechanism was what created liability.
But where the Moorhouse case found liability because the university had a degree of physical control over the copying environment, in the iiNet case, the High Court found no such control existed over a dispersed, user-directed internet service.
Another case was Telstra Corporation Limited v Australasian Performing Right Association Limited [1997] HCA 41. The High Court split four to three over whether Telstra had infringed copyright by playing recorded music to callers placed on hold. The majority held that Telstra had “broadcast” the work and therefore “performed” it in public, attracting royalty obligations. The minority argued Telstra was simply a carrier, transmitting a signal rather than performing a work.
Would generative AI pass the same test?
iiNet did not select, copy, or store the infringing content that passed through its network. The question now is whether AI companies can argue the same.
Training a large language model involves ingesting and processing vast quantities of copyrighted text, images, and other materials, then extracting patterns, relationships and structures from them.
The resulting model is shaped by that copyrighted material, and it provides outputs that can resemble (often closely) the substance and style of what it was trained on. On top of that, AI companies are charging subscription fees for access to these models built on copyrighted material.
“It kind of looks like they are providing the content and the means of infringement, and getting directly paid for it,” Malone said.
“If the BitTorrent software was regarded as the photocopier, the means of infringement, the software that the end-customer used, then I think there is an argument that the AI clients are a close parallel.
“This will be an interesting legal question for a future court.”
It is a question that courts in the United States, the United Kingdom and Europe are currently grappling with, as litigation brought by authors, publishers, musicians, and visual artists against AI developers accumulates.
An example is the recently settled class-action lawsuit against Anthropic. The company admitted to downloading millions of pirated, copyrighted books to train its AI models. The outcome of that lawsuit was that authors received about $3,000 per book.
Anna Funder was one of the many Australian authors who were party to the class-action lawsuit. Speaking at an event at Sydney Town Hall earlier in August, the Stasiland and All That I Am author said 17 editions of her books in different languages were fed into the company’s AI models.

Author Anna Funder signing books. Image credit: Stanny Angga via Wikimedia Commons.
She discussed what the AI industry's demand for training data looks like in practice.
“The LLMs were first trained on people's emails, on Reddit posts, on publicly available stuff on the internet,” Funder said. “What they wanted was...the material in books, which is better-written and edited and fact-checked and so on.”
As AI-generated content has proliferated online, the value of pre-AI human-authored material has increased, including for AI companies looking for high-quality material.
“What they particularly wanted were books published before 2022, because those ones are certain to be uncontaminated by AI,” Funder said.
Technical vs. legal protection
Some Australian rights-holders have explored technical mechanisms, like robots.txt directives and watermarking, to signal non-consent for AI training. The challenge is that these mechanisms are voluntary on the part of AI companies and unenforceable in the absence of legal protections.
Funder is also sceptical of how far technical protections can go.
“I don't think that's the answer. I think legal protection is the answer,” she said at the event.
“I think copyright has worked extremely well for 316 years since the Statute of Anne in 1710.
“Copyright goes hand in hand with democracy.”
Funder’s concern now is that “legal protections are getting outlawed”, an observation about the current direction of copyright policy in Australia.
She said instead of weakening copyright protections to accommodate AI companies, the government should tell those companies to obey Australia’s copyright laws and negotiate with creators if they want to operate here.
In Malone’s view, the path forward for a healthy media ecosystem in Australia is to pay creators.
“If the AI companies want to make relevant copyrighted content available to customers, such as news, someone has to pay for it,” he said.
“We all want a healthy media to exist and for content creators to be paid for their work.”
Part two explores how a recent legislative amendment may have dissolved the copyright protections that iiNet’s victory solidified.
