AI-Generated Content and Copyright Law in Australia: Who Owns What?

Australian businesses and creators are using AI tools to generate text, images, music, code, and video at a scale that would have been impossible two years ago. Most of them have not thought about who legally owns what the AI produces — or whether that output is protected at all. Here is the uncomfortable truth: under the Copyright Act 1968 (Cth) as it currently stands, content generated entirely by an AI system likely has no copyright protection in Australia. No protection means anyone can copy it, sell it, or publish it as their own. Meanwhile, the training data used to build AI models may itself infringe the copyright of the creators whose work was used without licence. Both sides of this equation have significant commercial and legal consequences — and the law has not yet caught up.

The Ownership Problem: Who Is the “Author”?

Copyright law everywhere in the world is built around a single foundational concept: authorship. Copyright protects the original expression of a human author’s creative effort. The Copyright Act 1968 defines copyright in terms of “works” created by “authors” — and the entire framework assumes the author is a human being.

AI systems have no legal personality. They are software. They cannot hold rights, they cannot enter contracts, and they cannot be authors in the legal sense. When an AI generates an image, a piece of music, or a legal brief, the AI is producing output through a process that involves no human creative decision-making at the moment of creation — the AI’s statistical models do the work.

This creates a fundamental gap: if there is no human author, there is no copyright. And if there is no copyright, the output is in the public domain — free for anyone to use, copy, or commercialise without permission or payment.

The question of how much human input is required to establish authorship — and therefore copyright — is the central unresolved legal question in Australia’s AI copyright debate. It is not a simple binary.

What the Copyright Act 1968 Currently Says

The Copyright Act 1968 has not been amended to address AI-generated content. The current position in Australia is derived from the Act’s existing provisions interpreted by courts and commentators in the AI context.

Training Data: Is Building an AI Model Infringement?

The most commercially significant copyright question in the AI space is not who owns the output — it is whether the inputs were lawful in the first place. Training large language models, image generators, and music AI systems requires enormous datasets of existing human-created work: books, articles, photographs, artwork, music recordings, code repositories.

In most cases, this training data was scraped from the internet without the knowledge or consent of the copyright owners. Whether this constitutes copyright infringement depends on two things:

1. Whether the training process involves reproduction of copyrighted works. Scraping and storing content in a training dataset almost certainly involves reproduction. Creating embeddings or model weights derived from copyrighted works may also constitute an adaptation or a derived work — though this is contested.
2. Whether any exception applies. In Australia, the relevant exceptions are the “fair dealing” provisions in sections 40–42 of the Copyright Act 1968. These cover research or study, criticism or review, reporting news, and judicial proceedings. None of these exceptions covers commercial AI training. Australia has no equivalent of the “text and data mining” exception that exists in the European Union (under the DSM Directive) or the broad “fair use” doctrine in the United States that AI developers have invoked.

The practical implication: Australian companies training AI models on copyrighted content — without licence — are likely infringing. Foreign AI companies whose models are accessed by Australian users may also face liability if their training involved Australian copyrighted works. Several class actions and individual infringement claims have been filed overseas (in the US and UK) against major AI developers on exactly this basis. Australia has not yet seen equivalent litigation, but it is a matter of time.

The Spectrum: When Does Human Input Create Copyright?

This is not a binary question — it is a spectrum. At one end: a user types “generate a poem about autumn” and accepts the first output verbatim. At the other: a writer spends hours refining an AI-assisted draft, making hundreds of deliberate creative choices about structure, voice, word choice, and theme. Between these extremes is a vast grey zone.

The principles Australian courts are likely to apply (extrapolated from existing copyright authorship cases) suggest:

• Purely AI-generated output without human creative input → no copyright. The act of prompting at a generic level is probably insufficient to establish authorship.
• Highly specific prompting that directs creative choices (style, structure, content, voice, perspective) → possibly some copyright protection for the prompt-derived expression, though the strength is uncertain.
• Significant human editing and creative decision-making applied to AI-generated material → the human’s contributions are likely protected, but only for the expression resulting from those contributions, not the underlying AI output.
• AI-generated elements incorporated into a predominantly human-authored work → the human-authored elements retain full copyright; the AI-generated elements within may not be protected if they can be identified and separated.

Practical Risks for Businesses Using AI

For businesses that have integrated AI tools into their workflows, the copyright uncertainty creates concrete commercial risks:

• Your AI-generated marketing materials, website copy, or product images may have no copyright protection. A competitor who copies them is not infringing any right you hold — because you hold none.
• AI output may incorporate third-party copyrighted material. Large language models and image generators have been found to reproduce portions of training data in their output. If an AI tool generates content that resembles or reproduces copyrighted material, and you publish or commercialise it, you may be the one facing an infringement claim — even if you had no knowledge of the underlying issue.
• Contracts assigning or licensing “AI-generated IP” may be void or unenforceable to the extent they purport to transfer rights that do not legally exist. A contract that assigns “all copyright in the content produced” cannot assign what the law does not recognise.
• Employment and freelance agreements need to address AI-generated content specifically — including questions of ownership, disclosure obligations, and quality standards.

AI Platform Terms of Service: What They Actually Say

Each major AI platform has its own terms of service on the question of ownership. These terms — which most users accept without reading — have significant implications:

• OpenAI (ChatGPT, DALL-E): OpenAI’s terms generally assign ownership of output to the user, to the extent permitted by law. They explicitly note that AI outputs may not be unique — identical prompts may generate similar content for different users. OpenAI retains a licence to use inputs for model improvement (adjustable in settings).
• Google (Gemini): Google’s terms allow users to retain ownership of their input and output but grant Google a broad licence to use the content for service improvement and other purposes.
• Midjourney: For free users, Midjourney content is licensed under Creative Commons (CC BY-NC 4.0 — non-commercial use only). Paid subscribers may use content commercially, but Midjourney retains a licence to display generated images.
• Adobe Firefly: Adobe specifically trained Firefly on licensed Adobe Stock content and public domain material — a notable attempt to address the training data infringement issue. Adobe’s commercial output is positioned as copyright-clean for licensed users.

Critically, none of these platform terms can create copyright protection under Australian law where the Copyright Act 1968 does not recognise it. Platform terms address ownership as between the platform and the user — they cannot override the statutory requirement for a human author.

What Reform Is Coming in Australia?

The Australian Government has acknowledged that the Copyright Act 1968 requires updating for the AI era. The Attorney-General’s Department and IP Australia have been consulting on reform since 2023. Key positions under consideration include:

• A “text and data mining” exception — permitting reproduction of copyrighted content for the purpose of training AI models on a non-commercial research basis, and potentially on a broader commercial basis subject to remuneration rights for copyright owners.
• Amended authorship provisions — potentially allowing copyright in computer-generated works to vest in the person who made the arrangements for the creation of the work (following the UK model under section 9(3) of the UK Copyright, Designs and Patents Act 1988), rather than requiring a human author in the traditional sense.
• Disclosure requirements — obligations on users or publishers of AI-generated content to disclose that the content was AI-generated, to protect both copyright owners and consumers.
• Moral rights and AI — clarification that AI systems cannot hold moral rights (right of attribution, right of integrity) and that human creators’ moral rights are not infringed by AI use of their works in training (controversial).

As at August 2026, no amending legislation has been introduced. The reform process is ongoing. Businesses and creators should monitor IP Australia and the Attorney-General’s Department for updates.

How to Protect Your AI-Assisted Work Now

Waiting for legislative reform is not a strategy if your business depends on AI-generated content. Practical steps to maximise your IP protection under the current regime:

• Document your human creative input. Keep records of your prompting process, the edits you made, the creative decisions you applied. This creates evidence of authorship for the human-creative-contribution elements of your work.
• Edit AI output substantially before publishing. The more substantially you alter, structure, and creatively decide on AI-generated material, the stronger your claim to copyright in the final work. A minimally-edited AI output has minimal copyright protection.
• Use trade secrets and confidentiality where copyright may not apply. Proprietary AI-generated business processes, systems, or models may be protectable as trade secrets or confidential information under contract — independent of copyright.
• Review AI platform terms before committing to a tool. Understand what rights you are granting to the platform, what rights the platform grants you, and whether the platform’s training data practices expose you to downstream infringement risk.
• Include AI provisions in your contracts. Employment agreements, freelance contracts, and IP assignment agreements should specifically address AI-generated content — who owns it, who is responsible for its accuracy, and what disclosure obligations apply.
• Seek legal advice on your specific use case. The copyright analysis differs between using AI for text, images, music, code, and data — and between different levels of human involvement. A one-size-fits-all approach misses the nuances.

Frequently Asked Questions

A Law Written Before the Internet, Applied to AI

The Copyright Act 1968 was not designed for a world where a machine can produce a commercially viable novel in minutes, or where an artist’s entire body of work can be ingested by a training algorithm overnight. The law will change — the question is when, and in which direction. In the meantime, Australian businesses and creators are operating with significant uncertainty on both sides of the AI content equation: the output may not be protected, and the inputs may not have been lawful.

The practical response to legal uncertainty is not to avoid the technology — it is to document carefully, contract clearly, and seek advice when the stakes are material. Copyright in the AI era is not just an academic question. It has direct consequences for who can profit from creative and commercial AI use, and who bears the cost when it goes wrong.

This article provides general information only and is not legal advice. The law governing AI-generated content and copyright in Australia is unsettled and subject to reform. Positions described reflect the current state of the law as at August 2026 and may change. For advice specific to your business, products, or creative work, consult a qualified intellectual property solicitor.