Australia EU US AI Copyright: Three Models Compared
Australia rejected opt-out. The EU adopted it. The US relies on fair use. Here’s what each approach means for your creative work and your rights.
ScriptShield Team
ScriptShield
We build tools that give screenwriters and creators cryptographic proof of authorship. Because your work deserves evidence as strong as your story.
Three of the world’s largest creative economies are handling AI and copyright in three fundamentally different ways. If you are a creator — a writer, musician, filmmaker, artist, designer or photographer — the jurisdiction you work in now determines whether you control your work’s relationship with AI or whether someone else already decided for you. ScriptShield tracks these developments because the evidence infrastructure creators need depends on which legal model prevails.
On 15 July 2026, Australia drew the hardest line yet. Prime Minister Anthony Albanese declared that no company should use Australian creative works for AI training without the creator’s control — including control of the price. He called anything less “theft.” That puts Australia ahead of both the European Union and the United States on creator protection. But it also raises the stakes for every creator who needs to understand what these three models actually mean in practice.
The Short Answer: Australia rejected the opt-out model and is exploring paid licensing. The EU lets AI companies train unless you actively opt out. The US treats training as potentially fair use. ScriptShield creates a tamper-evident record of your work, its versions and your permissions regardless of which jurisdiction applies. Start with a free hash receipt.
What is Australia’s approach to AI and creator rights?
Australia: Creator-Controlled Licensing
Australia has taken the most creator-protective position of any major economy.
In April 2026, the government rejected a text-and-data-mining exemption — the mechanism that would have allowed AI companies to train on Australian works without permission. The Attorney-General’s Department is now exploring a paid licensing model instead. Under that model, AI companies would need both permission and payment.
The PM’s speech went further. He did not merely say creators should be compensated. He said creators should control the price and value of their work. As Dr Mike Seymour from the University of Sydney observed, this position “is in stark contrast to the United States and appears to lean towards an even stronger position than that of the European Union.”
The Office of AI now sits inside the Department of Prime Minister and Cabinet. Australian Standards for AI are expected to be legislated early 2027. The CAIRG continues exploring three priorities: lawful licensing of copyright material for AI, certainty around AI-generated material and less costly enforcement including a possible small-claims forum.
The gap: detailed legislation has not been published. The mechanism for paid licensing has not been designed. Between now and early 2027, policy exists but enforcement does not.
What is the European Union’s approach?
European Union: Opt-Out With Teeth
The EU has the most developed regulatory framework but a weaker creator-protection model.
Under the Copyright Directive (Article 4), AI companies can train on copyrighted material unless the creator has explicitly opted out using machine-readable signals. The burden is on creators to act. If you do nothing, your work is available for training.
The EU AI Act adds transparency obligations. Providers of general-purpose AI models must publish summaries of training data, implement copyright compliance policies, designate complaints contacts and respect machine-readable rights reservations. Enforcement powers activated on 2 August 2026, with fines up to three per cent of global turnover.
In March 2026, the European Parliament voted to push further — calling for a European register of copyrighted works used in training, mandatory itemised disclosure of training data and the possibility that non-compliance could bar AI providers from operating in the EU.
The strength: actual regulation with actual enforcement and actual penalties. The weakness: it is an opt-out system. The default position is that your work is available unless you say otherwise. And the existing opt-out mechanisms — TDMRep, robots.txt, C2PA content credentials — are advisory signals that AI companies choose whether to respect.
What is the United States’ approach?
United States: Fair Use in the Courts
The US has no dedicated AI copyright legislation. It relies on fair use — a judicial doctrine that permits use of copyrighted material without permission where the use is sufficiently “transformative.”
AI companies argue that training constitutes transformative use because the model does not reproduce the original work — it learns patterns. Creators argue that training is commercial copying at industrial scale and that the resulting models directly compete with original work.
The courts are deciding. Over seventy AI copyright cases were active in the US as of mid-2026, including actions by authors, visual artists, musicians and news publishers against major AI companies. No Supreme Court ruling has settled the question.
The practical effect: US creators currently have no enforceable right to prevent AI training on their work unless they can win a court case. There is no opt-out mechanism, no licensing obligation and no government framework comparable to the EU or Australia. Individual creators bear the full cost and risk of litigation.
The gap is structural. Fair use was designed for a world where copying was occasional. AI training copies entire libraries. The doctrine is being stretched past its design parameters.
What does each model mean for independent creators?
Here is where it becomes personal.
If you are an Australian creator
The government has your back in principle. The PM’s speech is the strongest political commitment to creator control anywhere. But principle without enforcement infrastructure means your practical protection today is still the evidence you build yourself. The legislation is coming. Your record should be ready when it arrives.
If you are a European creator
You have legal rights but you must exercise them actively. Deploy TDMRep on your website. Block AI crawlers in robots.txt. Register with Spawning. Set C2PA training and data-mining assertions where your tools support them. If you do nothing, you have consented by default.
If you are a US creator
The legal landscape is uncertain. Fair use litigation takes years and costs more than most independent creators can afford. Building a provenance record does not change the legal framework, but it strengthens claims if a dispute ever reaches a courtroom or a future legislative change creates new protections.
If your work crosses borders
And most creative work does. An Australian novel sold through a US platform, reviewed by a European AI system and trained on by a company with servers in multiple jurisdictions faces all three regimes at once.
That is exactly why evidence matters regardless of jurisdiction. A timestamped version chain, an AI-assisted declaration, a documented permission record and a SHA-256 cryptographic hash work in every legal system because they establish facts, not rights. Facts travel across borders. The record you create today serves you wherever your work is challenged.
Side-by-Side Comparison
| Feature | Australia | European Union | United States |
|---|---|---|---|
| Default position | Permission required | Opt-out by creator | Fair-use question, no default protection |
| Licensing model | Exploring paid licensing | No mandatory licensing | Case-by-case litigation |
| Transparency obligations | Coming (2027) | EU AI Act — in force | None federal |
| Machine-readable opt-out | Under review | Legally recognised | Advisory only |
| Enforcement | Legislation expected early 2027 | Active — fines up to 3% of global turnover | Court proceedings only |
| Creator control of price | Explicitly recognised | Not addressed | Not addressed |
Which model will win?
Nobody knows. But the direction is clear.
The EU started with opt-out and is moving towards mandatory disclosure, itemised training-data lists and stronger enforcement. Australia skipped the opt-out model entirely and is heading for paid licensing. Even the US, despite its fair-use tradition, is seeing legislative proposals and major court decisions that will likely change the landscape within the next two years.
The trajectory is towards more creator control, not less. More transparency, not less. More documentation, not less.
The creators who build their evidence record now will be positioned for every model. The ones who wait will be starting from nothing when the rules arrive.
Ready for Any Jurisdiction
ScriptShield uses SHA-256 hashing and trusted timestamps to create a tamper-evident evidence record that works regardless of which legal framework applies.
Start With a Free Hash ReceiptProtect Your Creative Work
Generate SHA-256 authorship certificates and track who sees your scripts, manuscripts, and creative works.
Get StartedScriptShield provides evidence documentation tools for creators. It is not a law firm and does not provide legal advice.