Cullen International's latest global trend benchmark compared how copyright law applies to artificial intelligence (AI) across 14 jurisdictions around the world.
Key findings
The report's key findings include:
- Explicit AI training permissions remain rare: Only four of the 14 jurisdictions analysed explicitly allow the use of copyrighted material for AI training (the EU, Japan, Singapore and the US). Most others either prohibit it, are debating it, or have not addressed it in law.
- No jurisdiction has a fully operational, mandatory opt-out mechanism to prevent AI training from using copyrighted material: The EU offers the most structured framework through the text and data mining (TDM) exception in the Copyright Directive, which allows rights holders to reserve their rights. However, there is no agreed technical standard for machine-readable opt-outs. Most jurisdictions currently rely on other measures, including for example the terms and conditions available online.
- Compensation mechanisms for unauthorised use of copyright material for AI training are largely absent: No surveyed jurisdiction operates a specific compensation mechanism for the unauthorised use of copyrighted material in AI training. India proposed a hybrid model combining mandatory blanket licensing with a statutory remuneration right, but there is no timing for a decision.
- Human authorship is the dominant requirement for copyright protection of AI-generated content: Most jurisdictions require human authorship for copyright protection. Only Japan and Korea explicitly recognise some form of copyright protection for AI-generated content, though both still require a degree of human involvement. China has issued guidelines for court rulings to decide on copyright claims on AI-generated content, recognising AI-generated works gain copyright with original human input in prompts, parameters and post-editing.
For each jurisdiction, the benchmark covers whether:
- the use of copyrighted material in AI model training is allowed and if and how rights holders can opt out of their works being used for AI training;
- mechanisms exist to compensate rights holders for unauthorised use;
- AI-generated content is eligible for copyright protection.
The covered jurisdictions in the benchmark are: Australia, Brazil, Canada, China, the EU, India, Japan, Korea, Malaysia, Singapore, South Africa, Taiwan, the UK and the US.
For more information and to access the full report, please click on "Access the full content" – or on "Request Access", in case you are not subscribed to the Global Trends service.
See also our Global Trends benchmark on regulating artificial intelligence (AI).
more news
27 July 26
What is the state of implementation of the Cyber Resilience Act in the EU?
Cullen International's latest Benchmark analyses how 19 EU countries are preparing to implement the Cyber Resilience Act (CRA). It tracks national laws, designated authorities, incident notification bodies and penalties ahead of the regulation’s application dates.
13 July 26
Authorities around the world are addressing emerging regulatory topics in IoT
Our Quarterly Regulatory Update on IoT and M2M Services (Q2 2026) highlights how national regulators are shaping the future of IoT and M2M services in areas such as cross-border connectivity, device regulation, and security.
09 July 26
Global trends in AI regulation
Our latest Global Trends benchmark compared if and how artificial intelligence (AI) is being regulated across 14 jurisdictions around the world.