11-09-2026 
|
Getting your Trinity Audio player ready...
|
TL;DR: Generative AI models are trained on huge volumes of text, images, music, code, and other data, much of which is protected by copyright. The growing problem is not only deciding when AI training requires permission, but creating systems that show who owns the work, what uses are permitted, and whether rights exist or have changed. As AI consumes data at machine scale, copyright needs machine-readable infrastructure capable of operating at the same scale.
Key takeaways
- General-purpose AI models can require billions of data, including large volumes of copyrighted material.
- Creators often don’t know whether their work has been used for AI training, while developers rely on large datasets assembled from many different sources.
- Governments are increasingly pushing for more transparency around AI training data and copyright compliance.
- Machine-readable licensing could allow AI systems to identify ownership, permitted uses, rights reservations, and payment requirements before using data.
- Scalable public blockchains could provide a common record of ownership claims, licences, permissions, and other changes in rights.
It’s well understood that generative artificial intelligence (AI) needs data, and lots of it.
Text, images, code, books, articles, photographs, and other creative works have all become raw material (training data) for increasingly capable AI models. The problem is that most of this data already belongs to somebody.
The United Kingdom government’s 2026 report on copyright and AI says general-purpose foundational models can require billions of data points and will inevitably include copyrighted works obtained from multiple sources. It also states that developers provide varying levels of transparency about what they have used, while creators often don’t know whether their work was used in training datasets.
This creates a problem that goes well beyond deciding whether a particular AI model has infringed copyright—it raises questions about ownership, licensing, and permitted use. If AI systems are going to utilize creative works at massive scale, the systems managing these factors will need to operate at a similar scale.
Copyright was built around works, but AI operates on datasets
Copyright attaches rights to individual creative works. For example, a photographer owns rights to a photograph, an author owns rights to a book or article, and a musician or record label owns rights to a piece of music.
However, AI development works by training models on billions of data points gathered from public websites, archives, licensed collections, user data, private datasets, and third-party sources. The individual work becomes one data point in a set so large that neither the developer nor the creator may understand all of the rights attached to every input.
This makes copyright enforcement harder, but it also makes licensing harder. While a publisher can negotiate licensing for a collection of articles relatively simply, obtaining individual permissions for millions or billions of works is a much more complicated problem.
AI has exposed a mismatch between copyright operating at the level of individual works and AI development operating at the level of datasets. It has also shown that the traditional tools for dealing with these matters are obsolete. A new toolkit will be required.
The copyright problem is also a visibility problem
Much of the current debate is focused on what AI developers are legally entitled to use for training models. The answer depends on the jurisdiction, and there are many ongoing legal battles and debates on the subject.
However, before deciding whether permission is required, a more important question should be answered – can anyone reliably establish which works were used?
The U.K. government says more than 90% of respondents to its copyright and AI consultation supported disclosure of sources used in AI model training. Respondents cited copyright enforcement, licensing, respect for rights, and transparency as reasons for supporting disclosure. So, there’s broad public support for more transparency, at least in the U.K.
And the European Union has already moved in this direction—its AI Act means providers must comply with EU copyright law and publish sufficiently detailed summaries of content used in model training. It also requires model providers to identify and respect rights reservations made by copyright holders. The European Commission’s training-content template requires information about publicly available datasets, private datasets, scraped online content, user data, synthetic data, and other sources.
While even the EU’s regulations don’t provide creators with a list of every individual work used in model training, they illustrate the direction of travel. If copyright owners are expected to license and enforce rights while AI companies are expected to comply with those rights, both sides need better information about what data entered the system.
All of this means AI doesn’t only create a copyright problem. It creates a rights-information problem, too.
AI may need copyright it can read
One possible solution is making rights information machine-readable.
For example, imagine a photograph published online with a verifiable rights record containing information about its creator, copyright owner, permitted uses, commercial-use terms, AI training permission, attribution requirements, and more. An AI training pipeline could theoretically check those rights before adding the work to a dataset.
One publisher might allow indexing but prohibit model training while another might allow model training for a fee. It’s even possible to offer different types of licenses for training, retrieval, generation, or commercial use. Instead of rights existing only in licensing documents, website terms, databases, and separate legal agreements, some of that information could become machine-readable.
Currently, the lack of technical standards is one of the obstacles to such a market. Furthermore, the U.K. government’s 2026 impact assessment notes that existing protection measures aren’t always effective, that some web crawlers don’t comply with standards, and that there’s a lack of universally recognized tools for licensing.
A rights system for the AI era needs to give developers access to the data they need while giving creators control of their own work with the ability to charge for access to or use of it. It also needs to balance these needs without a lawyer to negotiate every machine interaction.
While models currently operate on a “scrape now and check rights later” process, a better solution would mean they check rights before the data enters the training system.
Data rights will need infrastructure
Making rights machine-readable still leaves one question unanswered: where does the authoritative rights information live? After all, copyright ownership can change, licenses can be granted or expire, rights can be transferred, and permission can apply to one use but not another.
Scalable public blockchains can provide part of the infrastructure for recording these changes. The creative work itself doesn’t need to be stored on the blockchain; a cryptographic identifier or hash can reference the work while transactions record ownership claims, permissions, licenses, rights transfers, and other changes. If any of these elements change, another transaction can update that state.
The same blockchain-based architecture could eventually support automated payments, too. If an AI system has permission to use a particular dataset or piece of information for an agreed price, the licensing and payment process can become part of the machine-readable workflow.
As always, this doesn’t solve every copyright question, and blockchains can’t decide who legally owns a disputed work, whether a particular use qualifies for an exception, or whether a specific use or output violates copyright agreements. These questions will still depend on laws, courts, and human intervention.
What scalable public blockchains can do is make the input side more transparent and traceable. With AI models already operating at machine scale, the rights infrastructure underneath the data will need to do the same.
FAQs:
Why does generative AI create copyright problems?
Generative AI models are trained using vast datasets which can include copyrighted texts, images, videos, music, code, and other creative works. Questions remain around when permission is needed, whether works were lawfully accessed, and how creators can identify and enforce the use of their material.
Do AI companies have to disclose their training data?
Requirements vary by jurisdiction. Under the EU AI Act, providers of general-purpose AI models must publish sufficiently detailed summaries of their training content and maintain policies for complying with EU copyright law.
What are machine-readable data rights?
Machine-readable rights are structured information that software can interpret automatically, such as who owns a work, whether AI training is permitted, what licenses apply, whether attribution is required, and what permissions currently exist.
Could blockchain tech help manage AI copyright licenses?
Scalable public blockchains could provide a shared record of ownership claims, license grants, transfers, permissions, and other rights information. The underlying copyrighted work and private license terms would not need to be stored on a public ledger.
Can blockchain determine whether AI infringes copyright?
No. Blockchains can provide records showing ownership claims, licensing rights, timestamps, and other information, but they can’t decide legal questions like whether a specific use was fair or infringed upon agreements.
In order for artificial intelligence (AI) to work right within the law and thrive in the face of growing challenges, it needs to integrate an enterprise blockchain system that ensures data input quality and ownership—allowing it to keep data safe while also guaranteeing the immutability of data. Check out CoinGeek’s coverage on this emerging tech to learn more why Enterprise blockchain will be the backbone of AI.
Watch: AI is becoming a game-changer for iGaming
