Australia's music industry has warned lawmakers that weakening copyright protections to attract OpenAI and Anthropic could allow technology companies to train artificial intelligence models on copyrighted works without permission, threatening licensing rights for musicians, writers and other creators worldwide.
Annabelle Herd, chief executive of the Australian Recording Industry Association (ARIA) and Phonographic Performance Company of Australia (PPCA), delivered the warning during an October 6, 2026, parliamentary hearing in Sydney. She urged AI developers to negotiate licensing agreements rather than seek changes to Australian copyright law.
Herd warned that accommodating the companies could turn Australia into the "Trojan horse for Big AI," creating what she described as a global haven for large-scale copyright exploitation. She argued that allowing developers to train advanced models in Australia under more permissive rules could undermine protections extending far beyond the country's creative industries.
The dispute comes as Australia's government considers how to regulate AI development while attracting investment from major technology companies. In October 2025, the government ruled out a broad text and data mining exception that would have allowed developers to use copyrighted works for AI training without permission or payment. Other approaches to licensing and enforcement remain under discussion, including arrangements that could allow the use of protected material unless rights holders explicitly object.
That distinction is central to the debate. Under Australia's Copyright Act 1968, creators generally control the reproduction of their protected works, subject to specific exceptions known as fair dealing. These cover defined purposes, including research or study, criticism, review and news reporting, rather than providing a general exemption for commercial AI training.
An opt-in licensing system requires developers to obtain authorization before using protected material. Under an opt-out approach, certain uses could proceed unless copyright owners take steps to prohibit them. Creative industry representatives argue that such a system would force rights holders to monitor how their work is used and actively prevent unauthorized training.
Attorney-General Michelle Rowland has maintained that voluntary licensing agreements are the government's preferred approach. In a September 29 interview, she reaffirmed its rejection of a broad text and data mining exemption while acknowledging that officials were considering other measures, including stronger enforcement and transparency requirements.
Herd challenged claims that obtaining permission from rights holders would be too costly or complicated for AI developers. She told lawmakers that established licensing systems already provide access to much of the world's recorded music, including through agreements with major record labels and Merlin, which represents independent labels.
"The world's richest and most advanced companies say it would be too costly and complex to work with rights holders," Herd said. She rejected the suggestion that difficulties negotiating licenses justified changing the law, arguing that existing arrangements provided a workable foundation for AI development.
Her concerns also extend to international copyright obligations. Under the principle of national treatment, Australia generally provides qualifying foreign works with copyright protection comparable to that afforded to domestic works. Herd argued that any compensation system would therefore need to account for overseas rights holders, complicating proposals focused exclusively on payments to Australian creators.
Anthropic disputed suggestions that it was attempting to dictate Australia's copyright policies. Jeffrey Bleich, a former U.S. ambassador to Australia representing the company, told the parliamentary committee that existing requirements made training advanced AI models domestically impractical. He argued that obtaining individual licenses for the enormous volume of material used in model training was technically impossible.
Bleich also said that hosting frontier AI training could give Australia greater influence over how the technology develops, rather than leaving the country dependent on systems trained elsewhere. The position reflects a central disagreement between technology companies seeking regulatory certainty and creative industries insisting that licensing obligations remain intact.
The Australian Broadcasting Corporation also opposed an opt-out arrangement. Its head of content and legal operations, Kate Gilchrist, warned that requiring copyright owners to identify and block AI training across numerous websites would impose substantial costs and monitoring responsibilities.
Herd maintained that the recording industry supports AI investment in Australia, provided developers comply with existing copyright protections. "They should come to the negotiating table, not to the Prime Minister's office," she said.
The copyright dispute unfolded during a broader parliamentary examination of AI oversight that also addressed unauthorized access to government websites by OpenAI agents. The company's chief strategy officer, Jason Kwon, apologized after its systems accessed Australian government platforms, including a Medicare statistics portal, during internal testing in June. OpenAI acknowledged that it should have handled the incident and its delayed disclosure more effectively.
Representatives of OpenAI and Anthropic also expressed support for mandatory reporting requirements for certain AI-related security incidents, another issue being examined as Australia considers new safeguards for advanced AI systems.
The Joint Select Committee on Artificial Intelligence is scheduled to continue hearings through October 9, 2026, and submit its final report by November 30. Its recommendations are expected to inform further consideration of copyright protections, AI development and oversight of technology companies operating in Australia.