AI and Intellectual Property: Copyright Issues for Australian Businesses

An Australian marketing agency uses an AI image generator to create product photography for client campaigns. Unknown to the agency, the generator was trained on copyrighted photographs without permission. A photographer sues the agency for copyright infringement. Who is liable? The AI vendor? The agency? The client? In Australia’s evolving AI and copyright landscape, the answer is: probably all of them, depending on the facts.

Intellectual property in the context of AI creates multiple, overlapping risks for Australian organisations. First: who owns the copyright in content generated by AI? Second: do organisations using AI tools that were trained on copyrighted material face liability for infringement? Third: what do the terms of service for commercial AI tools actually allow organisations to do with their outputs?

These questions sit in a legal grey zone in Australia. While the Copyright Act 1968 (Cth) is clear that human authorship is required for copyright protection, the Act predates generative AI. Courts and IP authorities have not yet definitively resolved ownership in the AI context. This uncertainty creates business risk.

Who Owns Copyright in AI-Generated Content?

Current Australian Law

Under the Copyright Act 1968 (Cth), copyright subsists in original literary, dramatic, musical, and artistic works created by a human author. The definition of “author” does not include artificial systems. If an AI model autonomously generates text or images, the Copyright Act is silent on ownership.

The practical implication: if a person directs an AI to generate content and creatively shapes, edits, or combines the output with human judgment, Australian courts would likely recognise that person as the author and copyright owner. If an AI operates autonomously with minimal human direction—for example, an algorithm that generates news articles with no editorial direction—copyright ownership is uncertain and possibly non-existent.

Australia differs from other jurisdictions. The US Copyright Office (2023) issued guidance that copyright registration requires human authorship, and works generated by AI without human creative input are not eligible for copyright protection. The UK, EU, and Canada are developing similar positions. Australia has not yet issued definitive guidance, but the trend is clear: copyright requires human authorship and creative direction.

Practical Implications for Organisations

For Australian businesses deploying AI to generate content, the safer legal position is:

Assume you own copyright only if you direct the AI and creatively shape the output. If you prompt an AI image generator with a detailed brief, review the output, edit it, and combine it with human-created content, you likely own the result. If you run an AI content generator with minimal direction and publish the raw output, copyright ownership is uncertain.

Document the creative direction and human involvement. Maintain records showing: the original brief, prompts provided to the AI, human review and editing, and creative decisions made. This documentation supports an argument that you, not the AI, are the author.

For internal use, assume you own the rights. Commercial AI tools (ChatGPT, Google Workspace AI, etc.) typically license outputs to you for broad use within your organisation. If you use the output internally, copyright ownership uncertainties are lower risk.

For external publication or commercialisation, seek legal advice or use AI tools with clear ownership terms. If publishing AI-generated content as your own creative work, or licensing it to third parties, IP ownership becomes material. Use tools with clear vendor representations about ownership rights.

Training Data Copyright Infringement: The Central Tension

Generative AI models are trained on vast datasets of text, images, code, and other copyrighted material, typically gathered without author permission. AI vendors argue this training constitutes “fair use” (or in Australia, “fair dealing”) under copyright law. Authors, publishers, and artists argue the opposite: training models on copyrighted work without permission is not fair use, and they are entitled to compensation.

This tension is playing out in global litigation. In the US, major cases are underway: New York Times v OpenAI, Sarah Silverman v Meta and OpenAI, and Getty Images v Stability AI. The outcomes will shape AI licensing for years. In Australia, no definitive case law yet exists, but the Copyright Act’s fair dealing exceptions are narrower than US fair use—Australian courts may provide less protection for training data use.

Risks for Organisations Using AI Tools Trained on Copyrighted Data

Organisations using commercial AI tools face indirect copyright exposure. If the underlying model was trained on copyrighted material, and the tool incorporates or regurgitates that material in outputs, does the using organisation face infringement liability?

Current legal thinking: the AI vendor likely bears primary liability for model training choices; the organisation using the tool bears secondary liability if it knowingly relies on infringing outputs. The risk escalates if the organisation knows (or should know) that AI outputs incorporate copyrighted material.

Example: A content agency uses ChatGPT to draft blog posts. If GPT outputs incorporate passages from published articles (verbatim or close paraphrasing), the agency publishing that content without independent verification faces copyright liability. The agency cannot defend by saying “the AI did it.”

For Australian organisations, the safer practice is: assume that commercial AI tools may have ingested copyrighted material in training, and verify that AI outputs do not incorporate copyrighted content before publication or commercialisation.

The New York Times Moment for Australian Businesses

The New York Times lawsuit against OpenAI (filed November 2023) alleges that ChatGPT was trained on NYT articles without permission, and can reproduce article content verbatim. The lawsuit is pending, but the implications are clear: if major AI models are trained on copyrighted material without licensing, publishers may demand compensation or obtain injunctive relief.

For Australian businesses, this creates a potential chain of liability. If your AI outputs infringe NYT copyrights, and the Times successfully obtains a judgment against OpenAI, the Times may pursue downstream users (like your business) for infringement. The safer position: verify your AI outputs, and if commercial AI tools are not adequately licensed, explore alternative tools or in-house models trained on licensed data.

Ownership and Licensing in Commercial AI Tools

When you use commercial AI tools (OpenAI’s ChatGPT Plus, Google’s Workspace AI, Microsoft Copilot, etc.), the service terms govern ownership and licensing of outputs. Understanding these terms is critical.

Typical Commercial AI Terms

Input data: Most commercial AI tools retain the right to use your inputs (prompts and context you provide) for model improvement and training, unless you explicitly opt out. This creates privacy and confidentiality risk: if you input sensitive business information, strategy, or customer data into a commercial AI tool, the vendor may use it for training.

Output ownership: Most tools grant you a licence to use outputs for personal, non-commercial purposes, or (for paid enterprise versions) for internal business purposes. Ownership of the output typically remains with the vendor; you receive a licence, not ownership.

Commercial use: If you want to commercialise AI-generated content (publish it, sell it, license it to third parties), check whether your tool licence permits this. Many free and standard commercial tiers prohibit commercial use of outputs.

Warranty and liability: AI vendors typically exclude warranties of accuracy, non-infringement, and fitness for purpose. Many exclude liability for third-party claims (copyright, patent infringement) arising from use of their tools. This creates organisational liability: if you publish AI output that infringes copyright, the vendor may not indemnify you.

Implications for Australian Organisations

Before using commercial AI tools for material business purposes, review the terms and understand:

• Whether your inputs may be used for training (if yes, and data is sensitive, use an enterprise version with opt-out)
• Whether outputs can be commercialised or published (if yes, under what restrictions)
• What warranties and indemnities the vendor provides (likely minimal)
• Whether the vendor licenses the underlying model on terms that permit you to rely on non-infringement

For sensitive or high-value work, use enterprise AI tools with negotiated terms, or build in-house models trained exclusively on licensed or original data.

Protecting Your Own IP When Using AI

Employment Contracts and IP Assignment

If your employees create AI-generated content as part of their employment, your employment contracts should clarify IP ownership. Standard employment contracts include IP assignment clauses (all work created in the course of employment belongs to the employer). Ensure these clauses are clear about content generated using AI tools, and that employees understand that outputs created on company time using company tools are company IP.

For freelancers or contractors creating AI-generated content, engage written contracts clarifying IP ownership, commercial use rights, and confidentiality obligations.

Marking and Metadata

For AI-generated content you publish or commercialise, include copyright notices and metadata marking your ownership: “(c) [Company Name] 2026. All rights reserved.” This signals your claimed ownership and creates evidence of intentional publication under your copyright claim.

Registry and Documentation

For valuable AI-generated works (logos, marketing materials, creative content), consider registering copyright or design marks with IP Australia or obtaining a design registration. While copyright exists upon creation (registration is not required), registration creates presumptive evidence of ownership that is valuable in infringement litigation.

Maintain documentation showing: the prompt or direction given to the AI, dates of creation, human editing and refinement, and your claimed authorship and ownership. This strengthens defensibility if IP ownership is later disputed.

Frequently Asked Questions

Who owns copyright in content generated by AI in Australia?

Under the Copyright Act 1968 (Cth), copyright ownership depends on whether the AI is a “tool” (you own the output) or an “author” (copyright may not subsist). Current Australian law does not recognise AI as a legal author or copyright owner. If a person directs the AI and creatively shapes the output, that person likely owns copyright. If the AI operates autonomously with minimal human direction, copyright ownership is uncertain.

What are the risks of copyright infringement in AI training data?

Organisations using commercial AI tools rely on models trained on copyrighted material without author permission. In the US, major lawsuits are challenging this. In Australia, the Copyright Act’s fair dealing exceptions are narrower than US fair use. Organisations using AI tools trained on copyrighted material face potential liability if those tools incorporate copyrighted content in outputs.

What do I need to know about commercial AI tool terms of service?

Most commercial AI tools retain ownership of the platform and model but grant users a licence to outputs. Key issues: many tools claim the right to use your inputs for model improvement (privacy/confidentiality risk); outputs may be used for research; and IP ownership in outputs varies. For sensitive work, check whether the tool permits opt-out from training data use and whether outputs are governed by the vendor’s IP terms or your own.

Key Takeaway

Copyright and IP in AI-generated content remains an uncertain area in Australian law. No definitive case law clarifies ownership in autonomous AI-generated outputs, and the ongoing global litigation over training data will shape how Australian courts eventually decide these questions.

In the interim, organisations should: assume you own copyright only if you direct the AI and creatively shape output; verify that AI-generated content does not infringe third-party copyright before publication; understand the IP terms in commercial tools you use; and document your creative direction and human involvement for content you claim to own.

Navigating AI and IP complexity? Anitech helps organisations understand copyright exposure, design AI IP strategies, and establish governance for safe, legally defensible use of generative AI. Contact us to develop your AI intellectual property framework.