AI Training and Copyright: Toward a Single Opt-Out Registry?
August 2, 2025, is fast approaching—a key date for the entry into force of the obligations set forth in the AI Act (EU Reg. 2024/1689). By then, providers of generative AI models will be required to ensure compliance with copyright law in text and data mining (TDM) activities, allowing rights holders to exercise their opt-out option.
But how is this opt-out exercised in practice?
Current technical solutions—such as the outdated robots.txt —fail to convince the cultural sector. More targeted alternatives, such as Spawning or ai.txt, are emerging, but the landscape remains fragmented. The recent ruling by the Hamburg Regional Court (Kneschke/LAION, September 27) recognized the validity of an opt-out expressed in natural language, reigniting the debate on how to interpret the regulation.
A single opt-out registry?
To bring some order to the situation, the European Commission is considering the creation of a central opt-out registry. It could be managed by the AI Office or the EUIPO, and would serve to provide a clear and accessible platform for those who wish to exclude their works from the training of AI models.
But doubts remain:
Who will cover the operating costs?
Isn't there a risk that this will become a bureaucratic hurdle for creators?
Why formalize a right that is already protected by law?
Innovation, yes, but not at the expense of copyright
Europe is betting on artificial intelligence, but it must do so while also protecting the work of authors, publishers, and content creators. The solution? Perhaps a simple, shared opt-out standard, without complications or gray areas.
The debate is on. Time is running out.