Article-At-A-Glance
- OpenAI’s senior vice president Ann O’Leary has made it clear: Australia must resolve its copyright barriers before OpenAI will consider hosting AI model training infrastructure on Australian soil.
- Prime Minister Anthony Albanese has committed to protecting creator rights, signaling that Australia will not introduce a copyright exemption for AI training — a position backed by the Attorney General.
- Australian copyright law currently contains no fair use exception, meaning AI companies cannot legally train models on protected content without permission and payment — a key distinction from U.S. law.
- Anthropic’s general counsel Jeff Bleich has separately confirmed that copyright resolution is a critical precondition for building AI data centres in Australia — showing this is an industry-wide pressure campaign, not just an OpenAI issue.
- The outcome of this standoff could shape how AI companies negotiate with governments globally — making Australia a critical legal test case worth watching closely.
OpenAI Just Threatened to Walk Away From Australia — Here’s What’s Actually at Stake
Australia is now ground zero for one of the most consequential copyright battles in AI history, and the legal implications stretch far beyond Canberra.
OpenAI’s senior vice president, Ann O’Leary, travelled to Australia’s capital and made a statement that reframed the entire debate: if the Albanese government maintains its current copyright protections, OpenAI will not establish AI model training operations in Australia. The offer on the table — becoming the first country outside the United States to host OpenAI’s AI training infrastructure — is significant. But the condition attached to it is even more significant for legal professionals tracking the intersection of intellectual property law and artificial intelligence.
This is not a casual negotiation. It is a direct challenge to a sovereign government’s legislative authority over copyright, dressed in the language of economic investment. For IP lawyers, tech regulators, and policy advisers, the structure of this ultimatum deserves careful unpacking.
What OpenAI Is Actually Demanding
OpenAI’s position, as stated through O’Leary’s comments in Canberra, is that Australia’s copyright framework creates a legal barrier to AI model training. What they are effectively requesting is a legislative carve-out — a copyright exception that would allow AI systems to train on protected works without obtaining a licence or making payment to rights holders. That is the core of the demand, even if it has not been framed that bluntly in public statements.
The Investment Leverage: Renewables and AI Data Centres
The leverage being applied is an economic one. O’Leary indicated that resolving copyright barriers would open the door for Australia to host AI model training — a significant infrastructure investment that would involve data centres, energy procurement, and long-term operational commitments. Anthropic’s general counsel Jeff Bleich echoed this, confirming that copyright was one of the critical issues that needed to be resolved before his company could build data centres in Australia. The implication is clear: change your laws, or lose the investment.
For legal professionals advising government or industry clients, this framing matters. It positions copyright reform not as a rights issue but as an economic development opportunity — a rhetorical strategy designed to create political pressure on legislators who might otherwise prioritise creator protections.
Why Australia’s Copyright Law Is the Sticking Point
Unlike the United States, which has a broad fair use doctrine under 17 U.S.C. § 107, Australia’s copyright framework under the Copyright Act 1968 (Cth) does not contain an equivalent general fair use exception. Australian law provides specific, enumerated fair dealing exceptions — for purposes such as research, criticism, news reporting, and parody — but none of these extend to commercial AI training on ingested copyrighted works at scale. This is the legal gap that OpenAI and Anthropic are pushing the Australian government to fill, and it is precisely the gap that the Albanese government has, so far, refused to create.
Albanese’s Position on AI Copyright
Prime Minister Anthony Albanese’s response to OpenAI’s ultimatum has been unambiguous. He has committed to copyright protections for creators in the AI age, stating that Australia’s laws will make the country’s position plain. That is a direct and deliberate signal to both the creative industries and to AI companies: the existing legal framework protecting rights holders is not going to be dismantled to accommodate AI training pipelines.
The Attorney General’s Ruling Against Copyright Exemptions for AI
The Attorney General’s position reinforces Albanese’s. Late last year, the Attorney General confirmed that there would be no copyright exception for AI training under the current government’s policy direction. This was not a tentative or exploratory statement — it was a definitive ruling out of a legislative pathway that AI companies had been lobbying heavily toward. For legal practitioners advising AI clients operating in or seeking to enter the Australian market, this confirmation narrows the viable compliance pathways considerably.
The Productivity Commission’s December Report Conclusion
Adding further institutional weight to the government’s stance, the Productivity Commission’s December report did not recommend introducing a broad fair use exception or an AI-specific training exemption. This matters because the Productivity Commission is the primary body through which major economic and regulatory reforms are assessed in Australia. Its conclusion effectively removes one of the key procedural avenues through which copyright reform could have been accelerated.
Albanese’s Direct Commitment to Creators
Albanese’s public language has been pointed. The phrase “our laws will spell that out plain as day” is not the language of a government open to negotiation on this specific point. It signals legislative intent and provides a clear marker for how Australian courts and regulators should interpret the government’s policy position in any future dispute involving AI training and copyright infringement claims.
What Australian Copyright Law Currently Says About AI Training
The Copyright Act 1968 (Cth) is the primary legislative instrument governing intellectual property protections in Australia, and it was written decades before generative AI existed as a concept. What this means in practice is that the Act’s protections apply fully to AI training scenarios — but its exceptions do not stretch to cover them. Any AI company ingesting Australian-hosted or Australian-owned copyrighted works for the purpose of training a machine learning model is, under the current legal framework, doing so without a lawful basis unless they have obtained a licence. For instance, companies expanding their AI data centers must navigate these legal challenges carefully.
This is not a grey area. The Act protects literary, artistic, musical, and dramatic works from reproduction without authorisation. Training a large language model or a generative AI system on those works constitutes reproduction at scale. Without an applicable exception or a licensing agreement in place, that activity exposes AI companies to significant infringement liability under Australian law. For instance, AI song generators could potentially face similar legal challenges if they use copyrighted music without proper licenses.
No Fair Use Exception Exists in Australian Law
Australia’s fair dealing provisions under the Copyright Act 1968 (Cth) are closed-category exceptions. They apply only to specific, defined purposes: research or study, criticism or review, parody or satire, reporting news, and professional legal advice. Commercial AI model training does not fall within any of these categories. There is no equivalent to the U.S. doctrine of transformative use that Australian courts could apply to reclassify AI training as a permitted activity. Rights holders in Australia therefore retain full control over whether and how their works are used in AI development pipelines.
How This Differs From U.S. Copyright Law
In the United States, the fair use doctrine under 17 U.S.C. § 107 allows courts to assess four factors when determining whether an unauthorised use of copyrighted material is permissible: the purpose and character of the use, the nature of the copyrighted work, the amount used, and the effect on the market for the original. The transformative use argument — that AI training repurposes content in a fundamentally new way — has been a central defence in U.S. litigation involving AI companies. That argument simply has no direct legal home in Australian law.
This distinction is precisely why OpenAI and Anthropic have flagged Australian copyright law as a barrier. The U.S. legal environment, while still contested, offers at least a plausible fair use defence. Australia offers none. For AI companies accustomed to operating under U.S. legal assumptions, the Australian framework requires an entirely different compliance posture — one built around licensing, not litigation strategy.
The Legal Implications for AI Companies Operating in Australia
Any AI company currently operating in Australia or using Australian content in its training datasets needs to take the current legal position seriously. The absence of a fair use exception is not a technicality — it is an enforceable legal reality that exposes these companies to copyright infringement claims that Australian courts are well-equipped to adjudicate.
Licensing Obligations Under Current Australian Law
Under the current framework, AI companies that wish to train models on Australian copyrighted content have one lawful pathway: licensing. This means identifying rights holders, negotiating terms, and executing agreements that authorise the use of protected works for AI training purposes. In practice, this could involve direct licensing with publishers, music labels, or collecting societies, or it could involve blanket licensing arrangements negotiated through bodies such as the Australasian Performing Right Association (APRA AMCOS) or the Copyright Agency.
What “Permission and Payment” Means in Practice
The phrase “permission and payment” has become shorthand for the position of Australian creative industry bodies: AI companies must obtain consent from rights holders and compensate them for the use of their works. This is not an abstract principle — it translates directly into licensing fees, royalty structures, and contractual obligations that AI companies would need to build into their operational and financial models before training on Australian content.
For legal practitioners advising AI clients, this means conducting thorough rights clearance audits before any training data is sourced from Australian markets. It also means advising clients on the reputational and financial risk of proceeding without clearance, particularly as Australian regulators and creative industry bodies are actively monitoring the space.
Deals Already Being Signed Globally: What Australian Law Requires
Globally, some AI companies have begun executing licensing agreements with content owners — a tacit acknowledgment that unlicensed training carries legal risk. News Corp, for example, has entered into content licensing arrangements with AI developers. These deals establish a commercial precedent that Australian law already demands domestically. What is voluntary in some jurisdictions is effectively mandatory in Australia given the absence of a fair use safety net. Legal advisers should be pointing to these global deals as the compliance benchmark for Australian operations.
How Creative Industry Bodies Are Responding
Australia’s creative sector has responded to the OpenAI ultimatum with a unified and legally grounded position. Industry bodies are not simply making moral arguments about creator rights — they are asserting enforceable legal entitlements under existing Australian copyright law, and they are doing so with the backing of a government that has publicly committed to maintaining those protections.
The response from collecting societies and representative organisations has been coordinated and consistent. The message being communicated to both government and AI companies is that the existing legal framework is sufficient, enforceable, and not subject to negotiation simply because a major technology company has attached an investment condition to its demands.
What is particularly significant from a legal strategy perspective is that these bodies are not waiting for litigation to establish their position. They are engaging proactively with the policy process, ensuring that any government consultation on copyright reform captures the full legal and commercial implications of introducing an AI training exemption.
“Australian copyright law already provides the framework. The question is whether AI companies will comply with it or continue to pressure governments to weaken it.”
— Representative position of Australian creative industry bodies in response to OpenAI’s ultimatum.
APRA AMCOS CEO Dean Ormston’s Position
APRA AMCOS CEO Dean Ormston has been explicit in asserting that AI companies must operate within the existing legal framework — obtaining permission and making payment before using music and associated works in AI training. His position is legally grounded: APRA AMCOS administers rights on behalf of its members under the Copyright Act 1968 (Cth), and those rights do not evaporate because a technology company argues that compliance is commercially inconvenient.
Ormston’s intervention is important for legal professionals because it signals that collecting societies are prepared to enforce their members’ rights actively. This is not a passive watch-and-wait posture. It is an assertive legal stance that creates real exposure for AI companies that have not secured the necessary licences before ingesting music, lyrics, or associated works into training datasets.
What AAM’s Response Signals for Licensing Negotiations
The Australian Associated Press and broader media body responses to OpenAI’s ultimatum signal that licensing negotiations in Australia will not follow the path of least resistance. Industry bodies are approaching these negotiations from a position of legal strength, not commercial desperation. The existing copyright framework gives rights holders the standing to demand terms — and the government’s public commitment to creator protections removes the political pressure that might otherwise push collecting societies toward unfavourable concessions.
For legal practitioners advising media clients or collecting societies, this is a significant tactical advantage. It means licensing negotiations with AI companies can proceed on the assumption that the alternative — unlicensed use — carries genuine legal risk for the AI company, not just reputational risk. That fundamentally shifts the bargaining dynamic in favour of rights holders.
Australia Is Not Alone: The Global Copyright vs. AI Battleground
Australia’s standoff with OpenAI is not occurring in isolation. Across multiple jurisdictions, the same tension between AI training practices and copyright law is playing out in courts, legislatures, and regulatory bodies simultaneously. In the United Kingdom, a government consultation on AI and copyright has been fiercely contested by the creative sector, with artists, publishers, and collecting societies pushing back against proposed text and data mining exceptions. In the European Union, the AI Act includes transparency obligations requiring AI developers to publish summaries of training data — a provision designed to facilitate rights enforcement without explicitly creating new exceptions.
In the United States, multiple active lawsuits are targeting AI companies over training data practices, including actions brought by The New York Times, a coalition of authors, and various music publishers. These cases are testing the boundaries of fair use in ways that will have direct implications for AI companies globally — including those operating in Australia. What makes Australia distinctive is that its legal framework already resolves the question that U.S. courts are still working through: without a fair use equivalent, the answer in Australia is clear. Unlicensed AI training on protected works is infringement.
Australia’s position may actually become a model for other common law jurisdictions that lack broad fair use doctrines — including Canada, New Zealand, and several Asian markets — where the legal default similarly favours rights holders over AI developers absent an explicit legislative exemption.
OpenAI’s Ultimatum Sets a Dangerous Precedent for AI Regulation
What OpenAI has effectively done in Australia is establish a template for how large AI companies can apply economic pressure to sovereign governments on matters of domestic law. The structure of the ultimatum — offer significant investment, attach a legal reform condition, frame it as an economic opportunity — is replicable across any jurisdiction where AI companies want legislative concessions. For legal professionals advising governments or regulatory bodies, this precedent deserves serious attention, especially in light of the latest developments in AI regulation.
If Australia capitulates, it signals to every other government that copyright law is a negotiable variable in AI investment deals. That would have cascading consequences for IP enforcement globally. Conversely, if Australia holds its position — as current political signals suggest — it establishes that democratic governments can resist this form of regulatory pressure while still engaging with the AI industry on commercially viable terms.
The legal principle at the centre of this dispute is fundamental: the right of a sovereign legislature to set and maintain the terms under which intellectual property is protected and exploited within its jurisdiction. OpenAI’s ultimatum does not just challenge Australian copyright law — it challenges the authority of any government to maintain IP protections in the face of commercial pressure from technology companies large enough to treat legislative reform as a dealmaking chip.
Frequently Asked Questions
The OpenAI copyright ultimatum has raised a number of precise legal questions among practitioners, policy advisers, and industry stakeholders. The following addresses the most critical of these questions based on the current state of Australian law and the publicly known facts of the dispute.
- Does Australian copyright law apply to AI training conducted offshore using Australian content?
- Can collecting societies initiate infringement proceedings without individual rights holder instructions?
- What damages are available under the Copyright Act 1968 (Cth) for large-scale AI training infringement?
- How does the government’s policy position affect pending or future copyright litigation?
- Are there interim compliance measures AI companies should implement while the policy debate continues?
Each of these questions has practical implications for how legal professionals structure advice to both AI company clients and rights holder clients in the current environment. The answers are grounded in existing statute and case law — not in the policy outcome of the OpenAI negotiation.
Can OpenAI legally train AI on Australian copyrighted content without permission?
No. Under the Copyright Act 1968 (Cth), reproducing copyrighted works without the authorisation of the rights holder constitutes infringement. There is no existing exception that applies to commercial AI training. Unless OpenAI has executed licensing agreements covering the specific works used in its training datasets, any training activity that incorporates Australian-held or Australian-owned copyrighted content without permission is an unlawful act under Australian law. This applies regardless of where the training physically occurs, as Australian copyright law can have extraterritorial reach where the rights holder is Australian and the infringing act produces effects in Australia.
What happens legally if OpenAI refuses to license content but continues operating in Australia?
Scenario: OpenAI declines to enter licensing agreements but continues to offer its services in Australia and/or uses Australian content in training datasets without authorisation.
| Legal Risk | Applicable Law | Potential Consequence |
|---|---|---|
| Copyright infringement (reproduction without licence) | Copyright Act 1968 (Cth), s. 36 | Civil liability; injunctions; account of profits or statutory damages |
| Authorisation liability (enabling infringement) | Copyright Act 1968 (Cth), s. 36(1A) | Secondary liability for platform-level infringement at scale |
| Regulatory enforcement action | ACCC oversight; potential future AI-specific regulation | Investigations, mandatory compliance orders, market access restrictions |
| Collecting society proceedings | APRA AMCOS, Copyright Agency mandates | Representative litigation on behalf of member rights holders |
The practical exposure is significant. Australian courts have jurisdiction to grant injunctive relief that could restrict OpenAI’s ability to operate services in the Australian market. Damages in copyright matters under Australian law can include an account of profits — meaning a court can order the infringing party to disgorge the profits generated from the infringing activity. At the scale of AI model training, that figure could be substantial.
Collecting societies like APRA AMCOS and the Copyright Agency have the legal standing and organisational capacity to pursue representative actions on behalf of large groups of rights holders, which removes the practical barrier of individual litigation that might otherwise limit enforcement. This is a material difference from other jurisdictions where enforcement is more atomised.
There is also the question of reputational and regulatory consequence. The Australian Competition and Consumer Commission has demonstrated a willingness to pursue large technology companies on matters of consumer and market regulation. While copyright enforcement sits primarily with rights holders rather than the ACCC, the broader regulatory climate in Australia is not favourable to AI companies that are seen to be flouting domestic legal obligations.
Legal advisers should be counselling AI clients that the cost-benefit analysis of non-compliance in Australia is unfavourable — particularly given the government’s public commitment to maintaining existing protections and the organised posture of collecting societies ready to act.
Does Australia have any fair use provisions that could benefit AI companies?
Australia does not have a general fair use doctrine. The Copyright Act 1968 (Cth) provides closed-category fair dealing exceptions that are exhaustively defined. None of the current exceptions — research or study, criticism or review, parody or satire, reporting of news, or provision of professional legal advice — are capable of being stretched to cover commercial AI model training at scale. Any argument to the contrary would face an extremely difficult path through Australian courts, and there is no current judicial authority supporting such an interpretation. The Productivity Commission has previously recommended introducing a broad fair use doctrine, but that recommendation has not been implemented, and the current government has not signalled any intention to act on it in the context of AI.
What legal framework governs AI training data licensing in Australia right now?
The governing framework is the Copyright Act 1968 (Cth) in its current, unamended form. There is no AI-specific legislation in force in Australia that creates new rights, exceptions, or obligations in relation to training data. This means that all existing intellectual property protections apply fully and without carve-outs to AI development activities. The framework requires rights holders’ authorisation before protected works are reproduced, communicated, or adapted — all of which can occur during the process of ingesting content into AI training pipelines.
Collecting societies operate under this framework with delegated authority from their members to license and enforce rights. The Copyright Agency licenses text and images on behalf of authors, publishers, and visual artists. APRA AMCOS licenses music and performing rights. Any AI company seeking a lawful pathway to use Australian content in training must engage with these bodies — or negotiate direct licences with individual rights holders — before proceeding. There is no regulatory sandbox, interim exemption, or transitional provision that suspends these obligations while the policy debate continues. For those interested in AI’s impact on music, here’s a comparison of AI music generators.
Could Albanese’s government face legal challenges if it changes copyright law to favour AI companies?
If the Albanese government were to reverse its position and introduce a copyright exemption for AI training, it would face several categories of legal and constitutional challenge. While the Commonwealth Parliament has broad legislative power over copyright under s. 51(xviii) of the Constitution, any amendment that reduces or extinguishes existing property rights without compensation could attract challenge under the just terms requirement in s. 51(xxxi), which protects against the acquisition of property on other than just terms.
- Rights holders could argue that a legislative exemption amounts to a compulsory acquisition of their intellectual property rights without compensation.
- Australia’s international treaty obligations under the Berne Convention and the TRIPS Agreement impose a three-step test for permissible copyright exceptions — any new AI exemption would need to survive scrutiny under this test.
- The Australia-United States Free Trade Agreement contains IP provisions that constrain Australia’s ability to reduce copyright protections below agreed minimum standards.
- Collecting societies and individual rights holders could pursue judicial review of any regulation made under a new exemption if the legislative drafting does not comply with existing treaty obligations.
The three-step test under the Berne Convention is particularly relevant. It requires that copyright exceptions be confined to certain special cases, that they do not conflict with normal exploitation of the work, and that they do not unreasonably prejudice the legitimate interests of the rights holder. A broad AI training exemption would face serious difficulty satisfying all three limbs of this test, particularly the third.
There is also the political dimension. Any government that introduced copyright exemptions benefiting foreign AI companies at the expense of domestic creators would face significant political and legal backlash from a well-organised creative sector that has already demonstrated its capacity to engage effectively with the policy process.
For legal professionals advising government clients on this question, the constitutional, treaty, and political constraints on any AI-friendly copyright reform are substantial. The current government’s stated position — maintaining existing protections — is not just politically motivated. It is the legally safer path, and the one most consistent with Australia’s existing international obligations.
