AI Strategy

Who Owns AI-Generated Content Copyright in Australia?

If a Midjourney render or a ChatGPT script sits in your next campaign, who actually owns it under Australian law? The short answer surprises most marketing teams: often, nobody does. This guide from Absolutely AI walks creators, in-house marketers and agencies through the 2026 position on AI copyright in Australia, what your vendor's terms really transfer, and how to paper AI-assisted work so ownership sticks.

A person mid-turn at a wide desk, one hand gesturing toward an open document and a laptop, framed three-quarter from the front in a minimal office

Australian copyright law was written for human authors, and in 2026 it still is. That single fact shapes every ownership question for AI-assisted creative work commissioned in this country, from a single-prompt product render to a fully art-directed campaign. For brands working with an AI creative agency like Absolutely AI, understanding where the human-authorship line falls is the difference between owning your campaign assets and holding nothing more than a contractual promise.

The short answer in 2026

Purely AI-generated output has no copyright owner in Australia. The Copyright Act 1968 (Cth) requires an identifiable human author who contributed independent intellectual effort to the material form of the work. Where a machine produces the output and a human merely pressed "generate", there is no author and therefore no copyright subsists at all. This puts Australia in a different position to the United States, where the Copyright Office allows protection over human-authored selection and arrangement, and to the United Kingdom, which has a specific computer-generated works provision assigning authorship to the person who made the arrangements necessary for the creation. Marketing teams commissioning AI work through a content creation partner need to plan around that gap, not assume the US or UK position applies here.

Why Australian law works this way

The human-authorship rule was crystallised in a run of Federal Court and High Court decisions that predate generative AI but map onto it cleanly. In IceTV v Nine Network (2009), the High Court reframed originality around independent intellectual effort directed at the material form of the work, not mere labour. Telstra v Phone Directories (2010) then applied that reasoning to hold that the White Pages and Yellow Pages had no copyright because their compilation was produced by an automated system with no identifiable human authors contributing the requisite effort. Acohs v Ucorp (2012) followed the same logic for computer-generated safety data sheets. The through-line matters: Australian courts have already decided, in three separate contexts, that machine-generated output without meaningful human authorship falls outside copyright. Generative AI simply presents the same question at higher resolution.

A person mid-step through a sparse studio corridor, holding a thin folder at their side, viewed from behind at a slight angle, coat mid-swing

When your prompt is (and isn't) enough

The practical question every creative team asks is where the human-authorship threshold actually sits for prompting, curation, editing and arrangement. There is no bright line yet in Australian case law, but the principles from IceTV give a defensible framework. A single-prompt Midjourney image, generated in one shot and used as-is, almost certainly attracts no copyright. The human has directed a topic but has not contributed independent intellectual effort to the material form: the pixels are the model's, not the prompter's.

Compare that with a heavily art-directed workflow: an initial concept sketch, multiple prompt iterations, deliberate selection between variants, compositing in Photoshop, colour grading, retouching and layout into a final campaign frame. Here the human contribution to the final material form is substantial and identifiable, and copyright can arguably subsist over the human-authored elements, even if the underlying generation does not. This is the working model behind serious AI brand photography and AI product photography engagements: the deliverable is not a raw render but a fully directed, composited and finished image where authorship is documented at every step.

Who owns it under the AI vendor's terms

Every major generative platform includes an ownership or assignment clause in its terms of service, and marketing teams routinely misread them. OpenAI's terms assign to the user "all its right, title and interest, if any, in and to Output". Anthropic's commercial terms take the same shape. Midjourney grants paid subscribers ownership of assets they create, subject to conditions. Adobe Firefly commercialises output through its indemnity program. Runway assigns output rights to the paying user. The critical phrase in almost every case is "if any". In Australia, if no copyright subsists in the output because there is no human author, there is nothing for the vendor to assign. You receive a contractual permission to use the output freely, but not an enforceable copyright you can license, assign onward, or sue a competitor for copying. For most in-house use this is fine. For a brand that needs to stop a competitor lifting its hero campaign image, it is not.

Employer, contractor and agency ownership

Where copyright does subsist, section 35 of the Copyright Act allocates first ownership. The default is that the author owns copyright. Two important exceptions: works made by an employee in the course of employment vest in the employer, and works commissioned in specific categories (photographs for private purposes, portraits, engravings) can vest in the commissioner. Freelancers and agencies do not fall into the employer exception, so client contracts must expressly assign copyright in writing under section 196 to transfer ownership. For AI-assisted deliverables, best-practice contracts now do three things at once: they identify the human contributors and their contribution, they assign whatever copyright subsists, and they grant a broad royalty-free licence over the balance so the client can use the work without worrying about which pixels crossed the authorship threshold. This is how a mature AI content agency engagement should be papered, and it is worth stress-testing your existing SOWs against it.

A clean two-panel document editor interface. Left panel shows a vertical checklist with labels: 'Human Contribution', 'Prompt Log', 'Edit History',

Risks beyond ownership

Ownership is only one layer of exposure. Training-data reproduction is the sharpest current risk: if a model regurgitates a substantial part of a copyright work in its training set, the user reproducing that output can be liable for infringement, regardless of whether they knew. Passing-off and Australian Consumer Law claims sit alongside that, especially where AI output mimics the style of a living artist closely enough to mislead consumers about endorsement or origin. Moral rights under Part IX of the Copyright Act, which cannot be assigned, continue to attach to any human-authored elements, so ghost-editing another creator's work into an AI pipeline without attribution is still a live risk. In December 2025 the Department of Industry, Science and Resources released updated guidance on "being clear about AI-generated content", pushing disclosure norms toward mandatory in some sectors and creating ACL exposure for brands that overstate the human origin of AI work. A well-run AI content studio should be able to walk you through each of these vectors as part of onboarding.

Practical checklist for Australian creators, marketers and agencies

The most useful thing a creative team can do right now is build an evidence trail that supports a human-authorship claim if one is ever challenged. The checklist below is what we recommend to clients running AI-assisted campaigns through our social ad creative and marketing workstreams.

  • Document human contribution. Keep a written brief, art-direction notes and a named creative lead for every deliverable.
  • Keep prompt and edit logs. Retain the prompt history, seed numbers, layered PSDs or project files, and version snapshots showing iteration.
  • Use contractual assignment. Every AI-assisted deliverable ships under a written section 196 assignment plus a broad backup licence.
  • Disclose AI use. Follow the DISR guidance and any sector-specific rules (broadcast, financial services, political advertising).
  • Avoid style-mimicking living artists. Style transfer prompts naming living artists are the fastest route to passing-off and ACL exposure.
  • Check vendor terms per project. Enterprise tiers usually offer stronger indemnities than consumer subscriptions.

Where Australian AI copyright law is heading

Reform is under active consideration. The Attorney-General's Copyright and AI Reference Group, convened in 2023 and running consultations through 2024 to 2026, has canvassed both a UK-style computer-generated works provision and a bespoke sui generis right for AI output. Neither has legislative form yet, and the Productivity Commission's parallel work on data and digital economy settings has pushed in the opposite direction, favouring caution before extending new rights. The realistic 2026 planning assumption for brands and agencies is that no reform lands before late 2027, and that the human-authorship rule continues to govern. Any contract, licence or brand-safety policy written now should be built on the current position, not the reform that might arrive.

Frequently Asked Questions

Can I copyright a Midjourney image I generated in Australia?

Not the raw output. If you have art-directed, edited, composited and arranged the image so that a human has contributed independent intellectual effort to the final material form, copyright can subsist over those human-authored elements.

Does OpenAI's assignment clause give me copyright over ChatGPT output?

It assigns whatever right, title and interest OpenAI has, which in Australia is often nothing because no copyright subsists in purely AI-generated text. You receive a strong contractual permission to use the output, not an enforceable Australian copyright.

Who owns AI content my agency delivers to me?

Whatever the contract says. Best practice is an express section 196 assignment of copyright plus a broad royalty-free licence covering any elements where copyright is uncertain. Check your SOW carries both.

Do I have to disclose that content is AI-generated?

The December 2025 DISR guidance pushes strongly toward disclosure, and sector-specific rules already require it in some contexts. Undisclosed AI use that misleads consumers can trigger Australian Consumer Law liability regardless of copyright.

Can I be sued if a model reproduces training data in my output?

Yes. Copyright infringement is generally strict liability in Australia, so reproducing a substantial part of a protected work exposes the user, not just the model provider. Enterprise indemnities from Adobe, OpenAI and Microsoft partially address this.

Do employees own AI content they create at work?

Where copyright subsists and the work is created in the course of employment, section 35(6) vests first ownership in the employer. Contractors do not fall within that exception and must assign in writing.

Is Australian AI copyright law about to change?

Reform is under consultation through the Attorney-General's Copyright and AI Reference Group, but no legislation is imminent. Plan on the current human-authorship rule continuing through at least 2027.

What is "independent intellectual effort" in practice?

It is intellectual effort directed at the material form of the work, established in IceTV v Nine Network. Choosing a topic is not enough; contributing meaningfully to the specific expression, structure or arrangement is.

Where this leaves brands and agencies

Australian AI copyright in 2026 rewards teams who treat generative tools as instruments in a directed creative process, not vending machines. Build the evidence trail, paper the contracts, disclose the workflow, and you own what can be owned and license what cannot. Absolutely AI works with in-house marketing teams and their legal counsel to build AI-assisted campaigns that hold up commercially and legally, and if you want to pressure-test your current setup, our creative agency team is a good place to start.

Ready to brief your next campaign?

Book a call