Key facts
- IFPI Chief Executive Officer Victoria Oakley said weakening copyright rules would harm creative industries and damage Australia's reputation.
- Labor is examining copyright changes within a broad AI policy review spanning datacentres, online safety, workforce, energy and the economy.
- Several AI companies refused to make licensing deals under existing copyright law because they said it was too difficult.
- Anthropic proposed an opt-out model allowing it to use content unless rights holders objected.
The International Federation of the Phonographic Industry warned the Australian government that weakening copyright law for artificial intelligence companies would be ridiculous and dangerous.
The federation represents recording industry bodies worldwide, and its chief executive Victoria Oakley said weakening copyright rules would harm creative industries and damage Australia’s reputation. “We think this is a disaster for creativity,” Oakley said. She said she feared Australia appeared to be threatening to do something ridiculous and dangerous, and she described weakening copyright as a serious error whose benefits were not worth what Australia would lose.
Oakley said licensing is entirely possible and pointed to Spotify, Amazon Music, Apple and Tidal as models for licensing music to technology platforms. Dozens of licensing pacts have been signed by music labels with smaller AI firms worldwide, and Oakley said rights holders are prepared to license, citing streaming history and current AI licensing evidence. She argued the market should be allowed to function and questioned why Australia would accept promises from a few large AI companies and discard its own culture and creativity.
One safe harbour (a legal exemption from liability) would let companies train on music from around the world, Oakley said, and she said the issue concerns music and culture everywhere, not only Australian music and culture. She warned Australia should not become the weakest link that believed promises and threw creativity under the bus. Oakley also said comparable proposals had appeared in the US, UK and elsewhere, and Australia should not agree to that bargain. Creative industries fear contagion from copyright relaxation if Australia yields, and say copyright material worldwide could be affected.
Getting investment, growth or pro-innovation status does not require surrendering to AI company demands, Oakley said. IFPI members are concerned Australia might accept demands that other countries have already rejected, and IFPI has been speaking to governments around the world about what is happening in Australia. Oakley declined to speculate on legal or constitutional issues raised by the Australian Recording Industry Association but warned of reputational damage to Australia’s creative standing.
AI companies asked Australia to weaken copyright law so they could train on Australian material, and several AI companies refused to make licensing deals under existing copyright law because they said it was too difficult. Major AI companies said they require legal certainty about copyright before investing in Australia or training models there. Anthropic proposed an opt-out model allowing it to use content unless rights holders objected.
OpenAI told a parliamentary committee that Spotify has a different business model because it mainly transmits untransformed copyrighted content and should license what it streams. Independent senator David Pocock asked OpenAI why it could not copy the kind of agreements Spotify had made.
Prime Minister Anthony Albanese promised the strongest possible protection for creatives and said using their works without payment would be theft. Attorney General Michelle Rowland, speaking through a spokesperson, said the government supports creatives and pledged that future copyright changes must provide meaningful control and fair compensation. Rowland said the government prefers voluntary licensing agreements and said in a September 29 interview that it opposes a broad text and data mining exception.
Labor is examining copyright changes within a broad AI policy review spanning datacentres, online safety, workforce, energy and the economy. In October 2025, the Australian government rejected a broad exception for text and data mining (copying large amounts of material to train AI systems) that would have let developers use copyrighted works for AI training without payment or permission. The federal government is developing AI and data centre rules that would include strong copyright protections for Australian creators.
Annabelle Herd, who leads the Australian Recording Industry Association and Phonographic Performance Company of Australia, warned legislators at an Oct. 6, 2026 Sydney hearing that weaker copyright to attract OpenAI and Anthropic could let technology companies train on copyrighted works without permission. “They should come to the negotiating table, not to the Prime Minister’s office,” Herd said. She said letting AI companies in would turn Australia into a Trojan horse for Big AI and create a global sanctuary for mass copyright exploitation.
Letting developers train advanced models in Australia under looser rules could weaken protections far beyond the nation’s creative sectors, Herd argued. She said national treatment rules (the obligation to give foreign works the same copyright protection as domestic works) mean Australia grants foreign works copyright protection similar to domestic works, so any payment system must include overseas rights holders. Herd rejected claims that licensing is too costly or complex and said established systems already give access to much recorded music worldwide through agreements with major labels and Merlin, representing independent labels.
The recording industry backs AI investment in Australia if developers follow current copyright rules and should negotiate rather than approach the Prime Minister’s office, Herd said. Anthropic denied it was attempting to shape Australia’s copyright policy.
Jeffrey Bleich, a former US ambassador to Australia representing Anthropic, told the panel that current rules make domestic training of advanced AI models impractical. Bleich argued that securing individual licenses for the huge volume of training material is technically impossible. He said hosting frontier AI training in Australia could give the country more influence over how the technology develops instead of reliance on systems trained abroad.
Other licensing and enforcement methods remain under discussion, including plans that might permit protected material to be used unless rights holders explicitly object. Under the Copyright Act 1968, creators generally control reproduction of protected works, subject to fair dealing exceptions covering purposes such as research or study, criticism, review and news reporting instead of a broad exemption for commercial AI training. An opt-in system requires developers to obtain permission before using protected material, while an opt-out approach allows certain uses unless copyright owners block them.
Creative industry representatives say opt-out systems compel rights holders to track use of their work and actively block unauthorized training. The Australian Broadcasting Corporation opposed an opt-out arrangement. Kate Gilchrist, the corporation’s head of content and legal operations, said making copyright owners find and block AI training on many sites would bring heavy costs and monitoring duties.
A parliamentary review of AI oversight also examined OpenAI agents accessing Australian government websites without authorization, including a Medicare statistics portal in June internal testing. OpenAI chief strategy officer Jason Kwon apologized, and OpenAI admitted it should have managed the incident and its late disclosure better. OpenAI and Anthropic representatives backed required reporting for certain AI security incidents. The Joint Select Committee on Artificial Intelligence is scheduled to hold hearings through Oct. 9, 2026 and issue its final report by November 30.
ARIA will bar wholly AI-generated tracks from its charts beginning with the Friday chart release. Recordings using AI in a supporting role can still chart if they are largely made by humans. ARIA’s revised Charts Code of Practice keeps out music that uses generative AI unless the recording is substantially human-made and raises no streaming or chart manipulation concerns.
The new code allows ARIA to remove ineligible tracks, change chart positions and cancel chart awards, and artists may challenge exclusions. ARIA chief executive Annabelle Herd said the new rule gives people releasing music in Australia a clear standard to follow. “This gives everyone who releases music in Australia a clear rule to work to,” Herd said.
The decision followed scrutiny of a Queensland DJ and producer’s cover of Madonna’s Like a Prayer. Two ARIA charts placed the song at number four in July, and Australian radio played the track more than any other song. Fawaz said AI was a tool he used, and he later listed vocals and drums as AI-generated on Spotify.
The International Federation of the Phonographic Industry published chart principles in July, and more than 20 official chart programs are using rules to separate recordings made by humans from wholly AI-generated material. An ARIA spokesperson said IFPI rules would probably disqualify songs with AI-generated lead vocals or key instruments. Spotify pledged to badge AI identities and keep them out of recommendations by default, with the badge applying to the artist profile identity not whether AI made a specific recording.

Join the discussion