OpenAI vice president of global policy Ann O'Leary told The Australian that the company "will not be able to build a … training centre here if the copyright law is as it exists". OpenAI's spokesperson added that the company does not currently train models in Australia and wants to engage "constructively" with government, creators and industry, according to B&T. The company is pitching compute investment as conditional on a change in copyright law.
The strongest case for holding the line
Before criticising the government's position, its logic deserves a fair statement. Creators argue that training is a commercial use of their work on a scale no individual licensor could negotiate against, and that a text-and-data-mining (TDM) exception would transfer value from Australian writers, musicians and publishers to foreign labs without consent or price. Prime Minister Anthony Albanese put it this way: no company should use Australian books, music, art or news to train AI "without the artist's control", including control of "the price and value of their work". That is a coherent property-rights argument, and it is the same one a pro-market publication should take seriously.
What the government has actually committed to
On 26 October 2025, Attorney-General Michelle Rowland said the government would not introduce a TDM exception to the Copyright Act 1968. She convened the Copyright and AI Reference Group (CAIRG) to look at alternatives, including a paid collective licensing framework, according to the Attorney-General's media release and ACS Information Age. By June 2026, the offices of Industry Minister Tim Ayres and Assistant Minister Andrew Charlton were still saying, per the ABC, "The government has ruled out a text and data mining exception. Our position has not changed."
The Department describes CAIRG as a standing consultative mechanism with no decision-making authority. That matters for OpenAI's timeline. Talks in a forum that cannot decide anything will not give a company planning a multi-year infrastructure build the certainty it wants.
Where the leaked options undercut both sides
The ABC reported on 15 September 2026 that Attorney-General's Department options put to rights-holders in early September would let AI firms train on works whose owners have not opted out. That would follow either payments to a central body or deals with a minimum number of rights-holders, per the ABC's report on the documents. One option reportedly states that "no additional $ would be payable to creators or rights holders beyond deals negotiated to achieve quota".
This is not the outright TDM exception the government ruled out, but it sits closer to one than the public rhetoric suggests. It also shows why OpenAI's framing should be treated as bargaining. The government is already working on options that would ease access, and the company is pushing for more of that easing.
The infrastructure claim needs scrutiny
The premise is that the copyright location of training determines where compute is built. That is plausible but not established. Training location matters legally because infringement analysis is territorial, so a model trained offshore does not trigger Australian copyright law at the training step. OpenAI's own spokesperson confirmed it does not train models in Australia today, which shows the company can operate around the constraint, and O'Leary herself said the issue is "really a copyright question more than a renewable energy question".
The economic stakes are real. In its August 2025 interim report, the Productivity Commission said AI will likely add more than $116 billion to Australian economic activity over the next decade, per its interim report page. Investment that does not arrive is a cost. But the Commission's broader argument was for a clear, predictable regime, not for any single outcome.
A proportionate path
The pro-innovation and pro-creator positions are not fully opposed. The costliest outcome for both is the current uncertainty: developers cannot price the risk, and creators cannot count on payment. A proportionate package would have four parts:
- Statutory clarity on training. Legislate a workable route, whether a collective licence or a narrow, remunerated permission, so investors know the rules before they commit capital.
- Transparency about inputs. Require disclosure of training sources at a level that lets rights-holders verify and enforce, without exposing trade secrets.
- Market-rate pricing, not a quota that pays nothing. A scheme where the price is set by negotiation and a collecting body, rather than by a quota that delivers no extra money, is more defensible to creators and less likely to trigger a backlash.
- A firm decision date. A reference group that advises without deciding is not a substitute for legislation.
Overseas developments show why this matters. EFF reported on 16 September 2026 that the Ninth Circuit rejected an attempt to stretch DMCA Section 1202 into a new source of liability in the GitHub Copilot litigation, in its analysis of the ruling. US courts are working through these disputes case by case. Australia can offer a statutory answer instead of years of litigation.
The bottom line
OpenAI's statement should not be read as a reason to abandon creators, and the government should not treat it as an idle threat. The sound response is to decide quickly on a licensing framework with real payment and real transparency, and to give investors a rule they can plan around.