Could AI companies scrape Australian content? A practical copyright guide
Australia is considering copyright reforms that could change how AI developers access online content. The proposal is not law yet—here is what it could mean.
What is being proposed?
The Australian Government is consulting on possible copyright reforms that could affect how artificial-intelligence companies access online material for training, testing or operating AI systems. The proposal is not final, so it does not currently create a new blanket right for companies to copy or scrape internet content.[S1]
The model under consideration could involve arrangements negotiated with representative groups for copyright owners. In practical terms, an AI developer might seek permission, pay for access or operate under a sector-wide framework rather than negotiate separately with every individual creator.[S1]
That would be a major change in how online content could be licensed for AI purposes. It would also raise difficult questions about who can represent creators, what content is covered, how payment is calculated and how an owner can object or opt out.[S1]
What does “scraping” mean?
Scraping is the automated collection of information from websites or other online services. A crawler may copy text, images, audio, video, metadata or other material and place it into a dataset.[S1]
Scraping and copyright are related but not identical issues. A website being publicly accessible does not, by itself, mean that its material can legally be copied, stored or used to train an AI model. The legal position may depend on the material, the purpose of copying, the licence terms, the scale of the use and whether an exception applies.[S1]
Privacy law can also apply. The Australian privacy regulator says publicly available or accessible data is not automatically lawful to use for generative-AI training, and web scraping may involve the collection of personal information.[S1]
What would change for creators?
Creators could gain a clearer route to licensing revenue if representative organisations are permitted to negotiate with AI companies. That could reduce the burden of pursuing individual agreements with large technology businesses.[S1]
The trade-off is control. A collective or representative arrangement may not reflect every creator’s preferred price, licence scope or objection. Creators should therefore examine any proposed rules about membership, representation, transparency, payment distribution, exclusions and withdrawal rights.[S1]
Until any reform is enacted, creators should:
- Keep dated records of original works, publication and ownership.
- Use clear copyright, licensing and AI-use notices where appropriate.
- Review contracts with publishers, employers, platforms and collecting societies.
- Preserve evidence of suspected copying, including URLs, screenshots and downloaded pages where lawful.
- Avoid assuming that a robots.txt file, paywall or terms page alone settles every copyright question.
A rights notice may help communicate an owner’s position, but it is not necessarily a complete legal prohibition or substitute for a contract.[S1]
What would change for websites and platforms?
Website operators may need to decide whether they want their material included in an AI licensing arrangement and whether they have authority to make that decision for contributors. A platform may host user-generated content without owning all underlying rights.[S1]
Operators should audit:
- Which content they own, license or merely host.
- Whether user agreements address AI training and automated copying.
- Whether commercial partners have separate rights or restrictions.
- How crawler access, authentication and rate limits are managed.
- What records exist showing consent, objections or licence status.
Platforms should also distinguish search indexing from model training. Both may involve automated access, but the legal and commercial permissions may differ.[S1]
What would change for AI developers?
AI developers would need to establish the legal basis for collecting and using training material rather than treating open access as unrestricted permission. A licensing framework could provide greater certainty, but compliance would still depend on the final legislation and the terms of each arrangement.[S1]
Developers should maintain dataset documentation covering source, date collected, rights status, licence terms, removal requests and downstream use. They should also apply filtering and deletion processes where material is outside the permitted scope.
Copyright clearance does not remove privacy obligations. The Privacy Act and Australian Privacy Principles can apply when personal information is collected, used or disclosed in connection with AI, including through scraping. The privacy regulator says organisations should collect personal information only when reasonably necessary and by lawful and fair means.[S1]
What should ordinary businesses do now?
Most businesses will not need to redesign their websites immediately. They should, however, identify where they rely on third-party content or AI tools.
Practical steps include:
1. Inventory text, images, software, recordings and databases used in business systems.
2. Check whether licences permit machine learning, automated analysis or onward disclosure.
3. Ask AI vendors how they obtain training data and handle customer inputs.
4. Do not upload confidential, personal or third-party copyright material without approval.
5. Keep procurement and legal records showing the checks performed.
6. Review contracts for warranties, indemnities, confidentiality and ownership of AI outputs.
The same material may trigger several legal regimes. A business can face copyright risk even where privacy risk is low, and privacy risk even where copyright ownership is clear.[S1]
Is anything legally settled?
No broad new permission has been established by the consultation proposal itself. It should be treated as a possible reform framework, not as enacted law.[S1]
The eventual rules could define eligible rights holders, authorised representatives, covered content, payment mechanisms, transparency duties, dispute resolution and exceptions. Until those details are settled, existing copyright, privacy, contract and platform rules remain relevant.[S1]
The safest approach is to separate three questions: who owns the material, what copying or access occurred, and what legal or contractual permission covers it. That analysis is more reliable than assuming either that all online content is protected from every form of AI use or that all public content is free to train on.