AI Copyright Lawyers Adelaide: Who Owns AI-Generated Content, Infringement Risk and AI Inventors
AI-generated content and copyright in South Australia
Content made with AI tools may belong to no one, and it can still infringe someone else’s copyright. Talon Legal advises Adelaide businesses, creators and agencies using generative AI for text, images, code and designs on who owns the result and how to manage the infringement risk. Book a same-day free 30-minute consultation or call (08) 7094 2021.
AI copyright key facts
| Issue | Position in South Australia |
|---|---|
| Human author needed | Copyright did not subsist in directories produced largely by automated processes (Telstra v Phone Directories, Perram J at [118] and Yates J at [169]). |
| Software-generated code | Source code created by a software system did not emanate from authors and was not an original work (Acohs v Ucorp at [57] to [58]). |
| How much human input | A question of fact and degree (Yates J at [169]). |
| Who owns the human part | The author (Copyright Act section 35(2)), or the employer for work made under a contract of service (section 35(6)). |
| Contractors and agencies | An assignment has no effect unless in writing signed by or on behalf of the assignor (section 196(3)). |
| Infringement | Doing or authorising an act comprised in the copyright without the owner's licence (section 36(1)), including in relation to a substantial part (section 14(1)). |
| Substantial part | Depends much more on quality than quantity (EMI v Larrikin, Emmett J at [52]). |
| Remedies | An injunction and either damages or an account of profits (section 115(2)). |
| Confidential material | Equity protects information identified with specificity, with the necessary quality of confidence, received in circumstances importing an obligation of confidence, against actual or threatened misuse without consent (Optus Networks v Telstra at [39]). |
| AI inventors | Only a natural person can be an inventor under the Patents Act and Regulations (Commissioner of Patents v Thaler at [113]). |
| First step | Same-day free 30-minute consultation. Call (08) 7094 2021. |
Need an AI copyright lawyer in Adelaide?
Our AI copyright lawyers in Adelaide act for businesses, creators and agencies using generative AI, and for owners whose work turns up in someone else’s AI output. The Copyright Act 1968 (Cth) decides whether AI-assisted material is protected and who owns it. The Patents Act 1990 (Cth) decides who can be an inventor. This page states the law current to 7 November 2026.
This guide is part of our artificial intelligence practice. Related guides cover AI contracts and data licensing, AI and privacy and technology IP and IT disputes.
Free Consultation. Before you publish, sell or register something made with AI, or reply to a copyright complaint, talk to an AI copyright lawyer. Book a same-day free 30-minute consultation or call (08) 7094 2021.
On this page
- Is AI-generated content protected by copyright?
- How much human input does copyright need?
- Who owns the parts a person contributes?
- What do AI tool terms say about ownership?
- Can AI-generated material be protected without copyright?
- Can AI-generated content infringe someone else's copyright?
- What about the data AI models are trained on?
- Can an AI system be named as an inventor?
- How should contracts deal with AI-generated content?
- How similar matters tend to play out
- How we act on AI copyright
- Case law
- Frequently asked questions
Is AI-generated content protected by copyright?
On the reasoning in the cases below, only to the extent a person authored it. Copyright subsists in an original unpublished literary work whose author was a qualified person when it was made (Copyright Act section 32(1)), and published works are covered on similar terms (section 32(2)). Literary works include computer programs (section 10(1)). Section 32 has no registration step. For AI material the real question is whether there is an author at all.
In Telstra Corporation Ltd v Phone Directories Company Pty Ltd [2010] FCAFC 149, the Full Court dismissed an appeal about directories compiled largely by automated processes, and copyright did not subsist in them.
Perram J noted that the Act does not deal expressly with software and authorship. In his Honour’s view, software is a tool, and the author of a computer-generated work will ordinarily be the person in control of it, so long as that person can be seen as directing or fashioning the material form. Where the operator does not control the material form, there is not enough independent intellectual effort to make that person the author: “a plane with its autopilot engaged is flying itself”. Where a computer performs functions ordinarily performed by human authors, copyright does not subsist ([118]).
Yates J reached the same result through originality. It is essential to originality that a work originates from an author or authors, and that is a question of fact and degree. The computer process was of such overwhelming significance to the directories’ material form that they could not be characterised as originating from authors ([169]).
In Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16, the claimed works were the HTML source code of documents a software system rendered ([50]). The Full Court agreed with the primary judge that the code was created by the operation of the system on material entered into it, did not emanate from authors, and was not an original work ([57] to [58]).
Neither case concerned a generative AI tool, and no case on this page decided how their reasoning applies to AI images, music or video. It is still the starting point. Material whose form a tool determined, rather than a person directing it, is unlikely to be protected, and a competitor may be free to copy it.
How much human input does copyright need?
Enough that a person, not the tool, directed or fashioned the material form (Perram J at [118]). Yates J called the question one of fact and degree ([169]). No case on this page drew the line for prompts. On Perram J’s reasoning, the effort that counts is effort directed at the material form of the work, so time spent refining instructions to a tool may not be enough on its own.
Output used exactly as a tool produced it, from a short prompt, is the most exposed, because the tool shaped its form. The more a person selects, edits, rearranges and rewrites, the stronger the claim to what that person contributed. Where a work mixes human writing and AI output, the claim is strongest for the human parts.
Because the question is one of fact, evidence decides it. Keep drafts, prompts, edit histories and version control records that show what people did. For code, record which parts came from an assistant and how a developer changed them.
Who owns the parts a person contributes?
The ordinary rules apply. Subject to the rest of section 35, the author of a literary, dramatic, musical or artistic work owns the copyright in it (section 35(2)). A work an employee makes under a contract of service belongs to the employer (section 35(6)), although an agreement can exclude or modify that rule (section 35(3)).
Staff using AI tools at work are in the same position. The employer owns what they author under their contract of service (section 35(6)), but that rule does not create copyright in what a tool generated. A staff policy on AI use, and on recording human input, protects the employer.
Contractors are different. Any copyright in what a freelance writer, designer, developer or agency creates for a business stays with it unless the contract assigns it, and an assignment has no effect unless it is in writing signed by or on behalf of the assignor (sections 35(2) and 196(3)). A signed agreement assigning future copyright vests it in the business as each work comes into existence (section 197(1)). Outside the commissioning rule in section 35(5) (photographs for a private or domestic purpose, portraits and engravings), payment alone does not move ownership.
An agency can only assign what it owns. If parts of a deliverable were generated with little human input, there may be no copyright in those parts to assign, however the clause is worded. The agency and PR agreements and website content and copyright guides cover those contracts.
What do AI tool terms say about ownership?
Whatever the vendor has written, and terms differ between tools and plans. They are a contract between the user and the vendor. They can settle who, between those two, may use inputs and outputs, but whether copyright subsists is decided under the Act (for literary works, section 32), so the terms may give the user nothing to own. Check:
- whether the vendor assigns, disclaims or keeps any rights in outputs;
- whether outputs may be used commercially on the plan the business is on;
- whether the vendor may use prompts, uploads and outputs to train its models;
- whether other users may receive the same or similar outputs, which affects exclusivity;
- whether the vendor will defend infringement claims about outputs, and on what conditions.
Terms can also restrict how outputs are used, for example by prohibiting their use to build a competing model. Those restrictions bind the user as a matter of contract whether or not any copyright subsists. The vendor’s privacy policy is worth reading too, because it says how prompts and uploads are handled. The AI contracts and data licensing guide covers negotiating those terms.
Can AI-generated material be protected without copyright?
Sometimes, through confidentiality and contract. Equity protects information that is identified with specificity, has the necessary quality of confidence, and was received in circumstances importing an obligation of confidence, against actual or threatened misuse without consent (Optus Networks Pty Ltd v Telstra Corporation Ltd [2010] FCAFC 21 at [39]). Unpublished outputs, prompts and workflows can meet that test while they are kept confidential. Material that has been published is unlikely to keep the necessary quality of confidence.
A contract can add to that protection, and a definition of confidential information does not by itself displace it. In Optus, an exhaustive contract definition of confidential information did not show an intention to exclude equitable obligations of confidence ([29]). Agreements with staff, contractors and clients should still name AI material, prompts and workflows as confidential, and limit what may be entered into tools that keep or train on inputs.
Can AI-generated content infringe someone else’s copyright?
Yes. Copyright in a work is infringed by a person who, not being the owner and without the owner’s licence, does in Australia, or authorises the doing in Australia of, any act comprised in the copyright (section 36(1)). For a literary work those acts include reproducing it in a material form, publishing it, communicating it to the public and adapting it (section 31(1)(a)). For an artistic work they include reproducing, publishing and communicating it (section 31(1)(b)). An act done in relation to a substantial part counts (section 14(1)).
A business that publishes AI output reproducing a substantial part of an existing work is exposed, because the publishing is its own act. The tool’s terms do not bind the copyright owner. Whether a particular AI output reproduces a work is a question no case on this page decided.
Substantial part is about quality. In EMI Songs Australia Pty Ltd v Larrikin Music Publishing Pty Ltd [2011] FCAFC 47, the question was whether recordings of “Down Under” reproduced a substantial part of “Kookaburra” ([1]). Emmett J said that depends much more on quality than quantity, and that an unoriginal part will not normally be a substantial part ([52]). Jagot J said the test is not whether the works are similar overall, but whether a substantial part of the copyright work was taken ([186]). Nicholas J said a part copied from another source, or commonplace, will not be a substantial part ([265]). The appeal was dismissed.
Emmett J also treated the idea of a four-part round as unprotected, and the specific melody expressing it as the work ([97]). For AI output, the risk sits in distinctive expression: a recognisable passage, melody, drawing, design or block of code can be enough, even if short.
A business that lets customers generate content on its own platform faces a further question: authorising infringement. The factors include the extent of its power to prevent the act, its relationship with the person who did it, and whether it took reasonable steps to prevent or avoid it, including complying with relevant industry codes of practice (section 36(1A)).
The owner can seek an injunction and either damages or an account of profits (section 115(2)). Additional damages can follow, having regard to matters including flagrancy and the need for deterrence (section 115(4)).
What about the data AI models are trained on?
For most businesses it is a risk to manage by contract and licence. The useful questions are who promises what about the material a model was trained on, and what others may do with the business’s own content.
A business using an AI tool should check what the vendor says about how its model was trained, whether the vendor will defend claims about outputs, and whether the business’s own uploads will be used for training.
A business whose content has value, such as a publisher, photographer or software house, should decide its position on training and state it in its licences, client agreements and website terms of use. Publishing and content agreements should say expressly whether the other party may use the work to train or fine-tune AI models.
A business building or fine-tuning its own model should source material under written licences that cover training expressly, keep records of sources, and take warranties from data suppliers. Personal information in training data raises separate rules, covered in the AI and privacy guide.
Can an AI system be named as an inventor?
No. In Commissioner of Patents v Thaler [2022] FCAFC 62, IP Australia had said a patent application did not comply with the Regulations because it did not identify a natural person as the inventor ([3]). The applicant had named DABUS as the inventor ([117]). A five-judge Full Court allowed the Commissioner’s appeal. The High Court refused special leave to appeal on 11 November 2022 (Thaler v Commissioner of Patents [2022] HCATrans 199).
A patent may be granted to the inventor or, under section 15(1)(b) to (d), to a person entitled to an assignment of it, a person deriving title, or a legal representative. The Full Court held that only a natural person can be an inventor for the purposes of the Patents Act and Regulations, and that such an inventor must be identified for anyone to be entitled to a grant under section 15(1)(b) to (d) ([113]). The Deputy Commissioner was right that naming DABUS as the inventor did not comply with the Regulations ([117]).
An invention developed with AI assistance therefore needs a human inventor who can be named. It must also meet the requirements in section 18(1), including being a manner of manufacture, novel, inventive and useful. Records of who conceived what, and employment and contractor terms dealing with inventions, matter. Whether a computer-implemented invention can be patented at all is covered in the technology IP and IT disputes guide.
How should contracts deal with AI-generated content?
By disclosure, warranties and indemnities, along the whole chain from tool vendor to agency to client.
- Disclosure: creators, contractors and agencies say where AI tools were used, and which ones.
- Warranties: deliverables are original or properly licensed, the tools’ terms allow the intended use, and confidential material was not entered into tools that train on it.
- Ownership: a written assignment signed by or on behalf of the creator (section 196(3)), covering future work (section 197(1)), with an acknowledgement that AI-generated parts may carry no copyright.
- Indemnities: for third-party infringement claims, with limits that match the risk.
Match the chain. A client’s warranty from an agency is only as good as the agency’s own rights under the tool’s terms, so the agency should check those terms before it promises anything. The AI contracts and data licensing guide covers the vendor contract. Where the risk is significant, screening deliverables for recognisable third-party material before acceptance is cheaper than an indemnity claim.
Authors and publishers dealing with AI-assisted manuscripts should read the book and e-book publishing agreements guide. Photographers, and businesses commissioning images, should read the photography terms and conditions guide.
How similar matters tend to play out
Illustrations of how the law on this page combines on ordinary facts, not predictions. Outcome depends on the documents, the evidence and the court.
Lower range. An Adelaide online retailer uses an AI tool to write its product descriptions, and a competitor copies them word for word. If no person directed or fashioned their form, the retailer may have no copyright to enforce (Telstra v Phone Directories, Perram J at [118]). Descriptions its staff substantially rewrote stand on firmer ground, and the drafts become its evidence on what is a question of fact and degree (Yates J at [169]).
Middle range. A design studio delivers a campaign mascot made with an image generator and assigns “all copyright” to its client. The owner of a well-known cartoon character says the mascot copies it. The client published the mascot, so it is exposed (section 36(1)), and a short but distinctive borrowing can be a substantial part (EMI v Larrikin, Emmett J at [52]). The client looks to the studio’s warranty and indemnity, and the studio looks to the tool vendor’s terms.
Upper range. A software start-up’s developers use an AI coding assistant for much of a product it sells under licence. In due diligence, an investor asks who owns the code. Parts were accepted from the assistant largely unchanged, one contractor never signed an assignment, and a scan finds a module closely matching third-party code the start-up has no licence for. The unchanged parts may not be original works (Acohs at [57] to [58]), any copyright in the contractor’s code stays with it (sections 35(2) and 196(3)), and the matching module risks an injunction (sections 36(1) and 115(2)).
How we act on AI copyright
Our AI copyright lawyers act for businesses, creators and agencies using AI, and for owners whose work has been copied. We start with what was made and by whom.
| Stage | What happens | What we do |
|---|---|---|
| Audit | The business needs to know what it owns in AI-assisted material. | Trace who made what, with which tools and under which agreements, and fix gaps with assignments. |
| Tool terms | The business chooses or reviews AI tools. | Check ownership, commercial use, training on inputs and indemnities. |
| Contracts | Contractors or agencies use AI on the business's work. | Draft disclosure, warranty, assignment and indemnity terms. |
| Complaint | A third party alleges copying, or the business's own work is copied. | Assess subsistence, ownership and substantial part, and respond. |
| Resolution | The dispute settles or goes to court. | Negotiate licences and settlements, or run the claim. |
We cannot promise an outcome, and no lawyer can. We can make sure the business knows what it owns before someone tests it.
Every hour matters. Book a same-day free 30-minute consultation, or call (08) 7094 2021 now.
AI copyright case law
No case on this page decided whether AI-generated content attracts copyright; Thaler was a patent case.
| Case | Court | Principle | Pinpoint |
|---|---|---|---|
| Telstra Corporation Ltd v Phone Directories Company Pty Ltd [2010] FCAFC 149 | Full Federal Court | Perram J and Yates J: works need a human author | [118], [169] |
| Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16 | Full Federal Court | Software-generated source code not original | [50], [57], [58] |
| EMI Songs Australia Pty Ltd v Larrikin Music Publishing Pty Ltd [2011] FCAFC 47 | Full Federal Court | Quality over quantity (Emmett J, Nicholas J); not overall similarity (Jagot J) | [52], [97], [98], [186], [265] |
| Commissioner of Patents v Thaler [2022] FCAFC 62 | Full Federal Court | Only a natural person can be an inventor | [3], [113], [117] |
| Optus Networks Pty Ltd v Telstra Corporation Ltd [2010] FCAFC 21 | Full Federal Court | Equity protects confidential information alongside contract | [29], [39] |
Case notes
Telstra Corporation Ltd v Phone Directories Company Pty Ltd [2010] FCAFC 149
The appeal was dismissed: copyright did not subsist in directories compiled largely by automated processes. Perram J said an operator who does not control the material form is not its author, and where a computer performs the functions of human authors, copyright does not subsist ([118]). Yates J said the computer process was of such overwhelming significance that the works could not be said to originate from authors ([169]).
Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16
The claimed works were the source code of documents a software system rendered ([50]). The Full Court agreed with the primary judge that the code did not emanate from authors and was not an original work ([57] to [58]).
EMI Songs Australia Pty Ltd v Larrikin Music Publishing Pty Ltd [2011] FCAFC 47
Emmett J held that both recordings of “Down Under” reproduced a substantial part of “Kookaburra” ([98]), and the appeal was dismissed. On substantiality, Emmett J looked to quality over quantity ([52]), Jagot J to whether a substantial part of the copyright work was taken rather than overall similarity ([186]), and Nicholas J said commonplace parts are not substantial ([265]).
Commissioner of Patents v Thaler [2022] FCAFC 62
A five-judge Full Court allowed the Commissioner’s appeal. Only a natural person can be an inventor for the purposes of the Patents Act and Regulations ([113]), so naming DABUS as the inventor did not comply with the Regulations ([117]).
Optus Networks Pty Ltd v Telstra Corporation Ltd [2010] FCAFC 21
The Full Court stated the elements of breach of confidence in equity ([39]), and held that an exhaustive contract definition of confidential information did not show an intention to exclude equitable obligations ([29]).
Frequently asked questions
Who owns AI-generated content in Australia?
Possibly no one. Copyright needs a human author, and it did not subsist in material produced by computer processes without one (Telstra v Phone Directories, Perram J at [118]; Acohs at [57] to [58]). What a person contributes belongs to its author or the author’s employer (Copyright Act section 35(2) and (6)).
Can I copyright an image made with AI?
Probably only to the extent a person directed or fashioned its form, on Perram J’s reasoning (Telstra v Phone Directories at [118]). No case on this page decided the point for images, and the answer turns on fact and degree (Yates J at [169]).
Is AI-generated code protected by copyright?
Code a person wrote or shaped can be; code a tool produced may not be. Source code created by the operation of a software system was not an original work (Acohs v Ucorp at [57] to [58]).
How much human input does copyright need?
Enough that a person, not the tool, directed or fashioned the material form (Telstra v Phone Directories, Perram J at [118]). It is a question of fact and degree (Yates J at [169]), so keep drafts and edit records.
Do AI tool terms give the user copyright in outputs?
Possibly not. The terms settle rights between the user and the vendor, but whether copyright subsists is decided under the Copyright Act (for literary works, section 32), so the terms may give the user nothing to own. They still matter for commercial use, training and indemnities.
Can using AI-generated content infringe copyright?
Yes, if the content reproduces a substantial part of an existing work and is copied or published without the owner’s licence (Copyright Act sections 14(1) and 36(1)). Substantiality depends much more on quality than quantity (EMI v Larrikin, Emmett J at [52]).
Who is liable if an AI output infringes copyright?
The person who does the infringing act, such as publishing the output, and anyone who authorises it (Copyright Act section 36(1)). Warranties and indemnities decide who bears the loss between the parties to a contract.
Can an AI be named as an inventor on a patent?
No. Only a natural person can be an inventor for the purposes of the Patents Act and Regulations (Commissioner of Patents v Thaler [2022] FCAFC 62 at [113]).
Does a freelancer own content they create with AI?
Any copyright in it stays with the freelancer unless assigned in writing, signed by or on the freelancer’s behalf (Copyright Act sections 35(2) and 196(3)). Parts generated with little human input may carry no copyright to assign.
Can a business keep AI prompts and outputs confidential?
Often, while they stay unpublished and are shared only in confidence. Equity protects information identified with specificity, with the necessary quality of confidence, received in circumstances importing an obligation of confidence, against actual or threatened misuse without consent (Optus Networks v Telstra at [39]).
Should a contract say whether work can be used to train AI?
Yes. Licences, publishing agreements and client contracts should say expressly whether the work may be used to train or fine-tune AI models, so the question is settled by agreement.
Do I need an AI copyright lawyer?
Before publishing, selling or registering material made with AI, or answering a copyright complaint, yes. An AI copyright lawyer can check ownership, the tool’s terms and the infringement risk. A free 30-minute consultation will tell you where you stand.