The future of tens of billions of dollars in AI investment in Australia is tied to a copyright law written before the moon landing. AI companies, including Anthropic, have made clear that major data centre projects are contingent on clarity of copyright settings, while Australian creators demand compensation for works already used to train generative models. The government is not expected to announce a decision on reform in the near term, but the legal standoff is now the central issue in Australia's AI policy debate.
A 1968 law meets generative AI
Australia's Copyright Act was drafted for an era of black-and-white television and radio. UNSW copyright law professor Kathy Bowrey said the system was already "incredibly complex and complicated" and "basically dysfunctional" before generative AI arrived. Training large language models involves collecting, copying, organising, testing, and reusing copyrighted material - each step potentially a separate infringement under Australian law.
Unlike the United States, Australia offers no fair use defence that courts can weigh for AI training. Professor Bowrey said it is "probable" that infringement has occurred at some stage of training, though no Australian court has yet heard a case. The jurisdictional challenge is significant: most major AI models are trained offshore, and litigation is expensive, limiting enforcement to large companies.
Investment and creator rights collide
Internal government documents obtained under freedom of information laws show that Anthropic, maker of Claude AI, identified Australia's stability, renewable energy potential, and close US ties as reasons to build data centres here. The Treasury briefing noted that any major investment would be "contingent on clarity of copyright settings". Tech and business councils have pushed for reform, arguing that copyright uncertainty blocks foreign investment without delivering income to creators.
For Australian artists, the reality of their work being used without permission is already confirmed. Grammy-nominated mixing engineer George Nicholas, formerly of electronic group Seekae, found 83 listings for Seekae tracks across four large song datasets used to train AI. "You knew that your IP was being trained on, but then to see it in the dataset … it really becomes real how much of your work exists within the training data of these models," he said.
Policy options and global implications
The government has several paths, ranging from leaving the law unchanged - as rights-holder groups urge - to collective licensing schemes or a permit system. A text-and-data-mining exception has been repeatedly ruled out. APRA AMCOS, which handles royalties for Australian and New Zealand artists, says no major AI platform has made a genuine attempt to negotiate with local rights holders. Collective licensing could simplify deals but raises difficult questions about representation, measurement, and whether payments reach individual creators.
The outcome will resonate beyond Australia. The US Copyright Alliance has argued that an Australian licensing deal could be cited in American courts as proof that agreements are workable. Both AI companies and rights holders view any local decision through the lens of a global tussle over training data, not just a domestic matter.
Why this matters for legal professionals
Copyright law is now the frontline of AI regulation, and the Australian debate is a microcosm of a global IP conflict. Legal professionals advising tech companies, publishers, or creators will need to understand the exposure from unauthorised training, the viability of licensing frameworks, and the strategic value of legislative clarity. For those building expertise in this area, AI for Legal Professionals Courses can provide the necessary grounding in the technology behind the legal questions. As policy shifts, the ability to interpret both the statute and the machine learning pipeline will be in high demand.
Your membership also unlocks:
