The global AI market is projected to skyrocket to over $1.3 trillion in the next five years. But beneath that value lies an unresolved conflict: the battle over generative AI and copyright as we know it today. AI models train on billions of works, so who should get paid? Who gets to call themselves a creator? The answer is being discussed right now in Europe.
The European Union Intellectual Property Office recently released a major study confirming what creators already feared – the existing legal framework is not really enough to deal with the scale of generative AI training.
The core problem is the EU’s Text and Data Mining (or the TDM) exception. It allows AI developers to use copyrighted content unless the rightsholder has opted out explicitly.
The EUIPO’s analysis shows an undeniable legal gap. The TDM opt-out system is undermined by some structural failures. First of all, there’s no single, harmonised, machine-readable standard for creators to reserve their rights across every platform. That lack of standardisation makes the opt-out practically unenforceable for individual creators against big-tech giants on a global scale.
Even more importantly, the study warns of a value gap that keeps growing. AI developers profit massively while creators receive little or no remuneration. In order to fix this, the EUIPO concludes that some sort of a compensation scheme would be essential.
This topic really frames the central policy debate over the new AI Act in the EU. Did the Act, by affirming the TDM exemption, actually grant Big Tech a free pass to extract data or did it strike balance that’s necessary for proper regulation?
The free pass argument is based on the following: Permission is granted unless you have the resources to opt-out. However, the AI Act introduces a safeguard: a mandatory transparency obligation. Providers of so-called General Purpose AI must publish a detailed summary of the copyrighted training data that was used. This doesn’t really stop the training, but it requires the data creators to prove a violation and enforce their opt-out rights, which is a step towards accountability.
Another question that many people, from lawyers to creators, are interested in is who is responsible for the output? Legally and philosophically, copyright protection should adhere to the principle of human authorship and originality.
In this view, the machine is nothing more than a really complex tool. A system’s properties, while they might take your breath away with their complexity, are basically results of human programming and data choice. Therefore, unassisted outputs should remain unprotected because they lack an original human creative spark, which is the backbone of creative content as such.
The human administrator who programs, guides, or refines the AI is the actual true author of the creative idea, and he therefore should hold the rights. This matters because this individual must also be recognised as the responsible party for the AI’s actions, whether it’s generating some new content or infringing on existing IP.
By prioritising this human-centric definition, policymakers will ensure the IP system continues to stimulate the human innovation, which was the thing that built the AI in the first place, rather than rewarding a machine.
The battle over AI and copyright defines the terms of trade for the next digital economy. The EU is attempting to govern a process of data mining which is now inherently global and invisible. The success of the AI Act will be measured by the enforceability of its transparency clauses and by whether it can sustain the fundamental promise of intellectual property that all creation, and all accountability, begins with the human being.
