Campaigners who helped force the UK to drop a plan allowing AI companies to train on copyrighted work without prior permission have warned Australia not to repeat the experiment. Only 3% of more than 10,000 respondents to the British consultation backed the opt-out model, prompting technology secretary Liz Kendall to withdraw government support in March. The dispute matters for business because the licensing rule chosen in Canberra will shape the legal risk of training models and deploying AI tools built on creative content.
Why the UK dropped the opt-out plan
Australia is weighing an opt-out system under which AI firms could use creators' work for training unless rights holders explicitly object. The proposal surfaced this month in leaked slides from the Attorney-General's Department, as reported by ABC News. The office of Attorney-General Michelle Rowland says it is still consulting on options intended to give creators meaningful control and fair payment. At the same time, OpenAI and Anthropic have urged Canberra to ease copyright rules, with OpenAI saying it could not establish an Australian training centre without legal changes.
Under opt-out, the default shifts to use unless refused, rather than a licence negotiated in advance. The UK version would have let developers ingest text, images and music at scale, placing the burden of monitoring and objection on authors and publishers. That design drew coordinated opposition led by film director and House of Lords member Beeban Kidron on the political front and by composer and former AI executive Ed Newton-Rex in public. Newton-Rex organised a silent album supported by more than 1,000 musicians and a book without text signed by 10,000 writers to show the value of the work at stake.
The British experience explains why Canberra faces pressure from both sides. Kidron told the Guardian, in reporting by Dan Milmo, that Prime Minister Anthony Albanese should not damage his standing by leaving Australian creatives unprotected. Newton-Rex argued the UK campaign convinced ministers that the public expects AI firms to pay for content. The technology industry reads the outcome differently: Antony Walker of trade group techUK said rights holders proved very effective at shutting down debate in Britain, while still insisting that model training and fair compensation can coexist.
What the dispute means for AI buyers
For companies buying or building AI systems, the immediate effect is legal uncertainty around training data. If Australia adopts opt-out, vendors could offer broader models faster and potentially at lower cost, since they would avoid clearing each work individually. If it moves toward explicit licensing, enterprise buyers would get clearer title to outputs but should expect slower data acquisition, narrower local models and higher fees passed through in contracts. Larger firms can manage licences and audits internally, while smaller firms will depend almost entirely on vendor warranties and documentation.
The current proposals leave several points unverified. Neither the leaked slides nor the consultation response define what counts as meaningful control, how an objection registry would work, or what fair payment would look like in practice. The news alone does not mean training on copyrighted material is now lawful in Australia, nor that UK rules are settled for good. Buyers should ask vendors where training data came from, how opt-outs are tracked across jurisdictions, what indemnity covers infringement claims, and how models would change if Canberra requires new licences.
A decision from Michelle Rowland's department on which copyright option moves forward, and whether OpenAI links an Australian training centre to that change, will show the direction of travel. A formal licensing framework or a retained opt-out would signal to vendors where to locate data work and how to price it. That choice will indicate whether other governments follow the British reversal or test a looser regime again.
