Australia Says No to AI Training Loophole: What the TDM Decision Means for Your Business

If you've been following the debate about AI and copyright, you've probably heard about "text and data mining" — or TDM. It's a technical term for something very simple: copying copyrighted material to train an AI model.

In October 2025, the Australian Government made a decision that places it at odds with many other countries: Australia will NOT allow AI companies to use copyrighted material for training without permission.

This might not sound like breaking news, but it's actually a significant policy choice with real consequences for businesses, creators, and the future of AI in Australia. Let's break it down.

What Is Text and Data Mining (TDM)?

Before we get into the policy, let's understand the technology.

When you train a generative AI model — like ChatGPT, Midjourney, or Claude — you need to feed it massive amounts of data. For a language model, that's text: books, articles, websites, code. For an image model, it's photographs, illustrations, paintings.

This process of copying and analysing copyrighted works to train an AI is called "text and data mining."

The question is: should this be allowed without asking permission from every copyright holder?

The Two Approaches

ApproachHow It WorksExample

TDM Exception

AI companies can copy and train on copyrighted material without permission

US, EU, Japan, UK

No TDM Exception

AI companies must get permission (or a licence) for every piece of copyrighted material

Australia, currently

Australia has chosen the second approach.

Why Did Australia Make This Choice?

The decision didn't come out of nowhere. Here's the context:

1. Backlash from Creators

The Australian Government consulted with artists, writers, musicians, and other creators through a process called the Copyright and Artificial Intelligence Reference Group (CAIRG). The response was overwhelmingly negative.

Artists were furious at the idea of AI companies using their work — often without compensation — to build products that could compete with them. The political pressure was significant.

2. Protecting Australian Creative Industries

Australia has a vibrant creative sector: film, music, publishing, advertising, and design. The government wants to ensure these industries aren't undermined by AI companies that can replicate creative work without paying for the original material.

3. Different Philosophy from the US

The United States has a "fair use" doctrine that courts have interpreted to allow AI training. But Australia doesn't have fair use. Our copyright system is more rights-holder focused, and this decision reflects that philosophy.

What This Means for AI Companies

1. You Need Permission (or a Licence)

If you're building an AI model in Australia, you can't just scrape the internet for training data. You need to:

Get licences from copyright holders for the material you want to use
Use only public domain material (works where copyright has expired)
Use material that is explicitly licensed for AI training (like some open-source datasets)
Create your own data (expensive and time-consuming)

2. Costs Are Going Up

Licensing copyrighted material costs money. Many AI companies in the US trained their models on data they didn't pay for. Australian AI companies will face higher development costs.

3. Competitive Disadvantage?

This is the big concern. If Australian AI companies have to pay for training data while US and European competitors don't, Australian AI companies will be at a cost disadvantage. This could slow down innovation and make it harder for Australian startups to compete globally.

4. Where Will the Data Come From?

AI companies will need to find alternative data sources:

Licensed datasets from publishers, stock photo agencies, and content platforms
Publicly available data that isn't copyrighted
Synthetic data generated by existing AI systems
User-generated content with appropriate licences

What This Means for Content Creators

1. You Have More Control

If you're a writer, photographer, musician, or artist, this decision gives you more control over your work. AI companies can't just use your material without your permission.

2. New Revenue Opportunities

You can now license your work to AI companies on terms you control. This could create new revenue streams:

Direct licensing deals with AI companies
Collective licensing through industry bodies (more on this below)
Royalty arrangements based on how your work is used

3. The Risk of Unauthorised Use

Despite the law, AI companies may still use copyrighted material without permission. The question is: will they be able to enforce their rights? Enforcement is always harder than prevention, and AI training happens at such scale that monitoring is extremely difficult.

4. Collective Licensing May Be the Answer

The Australian Government has said it will continue exploring collective licensing models. This is similar to how music royalties work in Australia:

A central body (like APRA AMCOS for music) negotiates licences on behalf of rights holders
AI companies pay into a fund
Rights holders receive compensation based on how their work is used

This could be a win-win: AI companies get access to data, and creators get paid.

What This Means for Other Businesses

1. Companies Using AI Tools

If your business uses AI tools like ChatGPT, Midjourney, or Claude, this decision doesn't directly affect you. You're not training models — you're using them. However, it may affect:

The quality of AI tools available in Australia (if training data is more limited)
The availability of Australian-trained AI models (which may be less competitive)
The legal risks of using AI-generated content (if the underlying training was unauthorised)

2. Companies in Creative Industries

If you're in advertising, marketing, design, or content creation, this decision protects your business. AI companies can't just replicate your creative work without permission.

3. Companies in Technology

If you're building AI products, you need to factor in the cost and complexity of obtaining training data licences. This may affect:

Your product timeline (licensing negotiations take time)
Your product cost (licensing fees add to development costs)
Your product features (limited training data may limit what your AI can do)

How Does Australia Compare to Other Countries?

CountryTDM ExceptionAI Training on Copyrighted Material

Australia

❌ No

Must have permission

United States

❌ No formal exception, but fair use applies

Generally allowed under fair use

European Union

✅ Yes (with opt-out)

Allowed unless rights holder opts out

United Kingdom

✅ Yes (non-commercial); commercial allowed unless opted out

Allowed with conditions

Japan

✅ Yes

Allowed

Canada

❌ No formal exception

Uncertain, likely requires permission

Australia is now one of the few major developed economies without a TDM exception. This is a significant policy divergence.

What Could Happen Next?

1. First AI Training Cases in Australian Courts

We haven't seen an Australian court case about AI training yet, but it's coming. When it does, the courts will have to decide:

Does using copyrighted material to train an AI model infringe copyright?
Are there any defences available?
What remedies should be available to rights holders?

2. Legislative Change Is Possible

The Government's decision today doesn't mean it's permanent. If the AI industry makes a strong case that the current approach is harming innovation, the government could reconsider. Policy can change.

3. International Pressure

If Australian AI companies struggle to compete because of the TDM restriction, there may be pressure to change the policy. Other countries are moving in the opposite direction.

4. Collective Licensing Develops

The Government has said it will explore collective licensing. If this develops successfully, it could provide a practical solution that satisfies both AI companies and creators.

What Should Your Business Do Now?

If You're an AI Company or Developer

Audit your training data. What copyrighted material are you using? Do you have licences?
Start negotiating licences now. Don't wait until you're sued.
Consider alternative data sources. Public domain, synthetic data, user-generated content.
Engage with creators. Build relationships with content providers before you need them.
Monitor policy developments. The government may revisit this decision.

If You're a Content Creator or Rights Holder

Understand your rights. You have control over how your work is used for AI training.
Consider licensing your work. There may be money to be made.
Document your work. Keep records of creation dates and ownership.
Join industry bodies. Collective licensing may offer better deals than individual negotiations.
Monitor for unauthorised use. AI companies may still use your work without permission.

If You're Using AI Tools in Your Business

Understand the risks. AI-generated content may be based on unauthorised training data.
Check your AI tool's licence. What are they allowed to do with your data?
Be cautious with sensitive content. Don't feed confidential or proprietary information into AI tools.
Stay informed. This area is evolving rapidly.

The Bottom Line

Australia's decision to reject a TDM exception is a clear statement: creators' rights come first. AI companies can't just take copyrighted material without permission.

But this decision also creates tension. It protects creators but may hinder AI innovation. It gives rights holders control but may put Australian AI companies at a competitive disadvantage.

The challenge going forward is finding a balance — a system that protects creators while enabling innovation. Collective licensing, voluntary agreements, and thoughtful regulation may be the answer. But for now, the message is clear: if you want to use copyrighted material to train AI, you need permission.

This article is for general information only and does not constitute legal advice. If you have questions about AI and copyright for your business, consult a qualified intellectual property lawyer.