Initiatives
The European Commission runs open consultations with interested parties on the future of EU regulation. We have taken part in a series of these initiatives.
1. A Single System for Labelling AI-Generated Content in the EU
Over the summer of 2026, YouTube, Instagram and Claude began marking AI-generated content in a special way. This was not a coincidence — it followed the EU AI Act becoming applicable.
Yet the label itself does not answer the main question — where did this content come from? We do not know which model was used to create it and what exactly the AI did to the content.
Our vision
To address this problem, CLAIMS took part in the European Commission's public consultation on the transparency of AI systems and proposed its own system — ETL-AI (EU Transparency License for AI-Generated Content). The CLAIMS proposal is available here.
We proposed:
- one clear approach to labelling across the EU — so that each service does not invent its own signs;
- showing not only that AI was used, but what exactly the AI did to the content — whether the content was fully generated by AI, is a deepfake, or was edited by a human;
- adding information about the origin of the content to metadata — including the ability to verify its source technically;
- setting common rules for the placement and visibility of labels — for text, images, video and audio;
- creating a public verification tool — through which users could check the origin of content;
- protecting labels from removal — and preserving them when content is redistributed or modified.
What changed in the final Code
The European Commission separately published the results of the public consultation and stated that the feedback received was used in preparing the final Code of Practice on marking and labelling AI-generated content.
| What CLAIMS proposed | What changed in the Code | What we got |
|---|---|---|
| One clear sign for the whole EU. We argued against a situation where every service invents its own way of labelling AI-generated content. | Already in the second draft of the Code, the authors proposed a common, free European label. The final version introduced the official “AI GENERATED” and “AI MODIFIED” icons. | The idea of a common European label made it into the final Code. The Commission, however, developed its own design instead of the system we proposed. |
| Show not only that AI was used, but what exactly the AI did to the content. | The final Code introduced two different labels: one for content fully generated by AI and another for content modified using AI. It also provides for further development of interactive labelling that could give more information about the modifications made. | The label became more informative: users can distinguish fully AI-generated content from content modified with AI. But the label still does not show the full history of where the content came from. |
| Set common rules for where a label should appear and how long it should remain visible. In ETL-AI, we even proposed specific rules for images, video, text and audio. | The final Code introduced a separate set of placement rules for labels across different types of content. | The principle of common rules was preserved. The Commission did not use our specific sizes, colours or timing intervals. |
| Create one clear tool for checking the origin of content, instead of making users find a separate verification tool for every AI provider. | The second draft of the Code contained a very similar idea — a common EU-wide entry point through which the verification systems of different providers could be accessed. In the final Code, the approach became more flexible: different interoperable tools and shared verification systems are allowed. | The problem was recognised, but a single European verification tool never appeared. |
| Do not try to solve the transparency problem by producing even more instructions and documents. | In the first draft of the Code, user documentation and training were framed more strictly. In the final version, this part became voluntary. | The Code became less bureaucratic in this respect. But we cannot claim that this happened specifically because of the CLAIMS comment. |
Outcome of the initiative
Labelling has arrived, but it still does not give us the full picture of where content came from — although now you can find out whether your consultant used Claude to write a text. You can, right?
2. A Single EU Registry for AI Training Opt-Outs
Same old song plays in the media whenever AI is trained on copyrighted materials: Sony Music Publishing and Warner Chappell filed a lawsuit against Anthropic. Courts and multi-billion-dollar settlements make headlines, but what if we focused on a solution that would actually work in an A2A reality?
The EU already has an answer taking shape. For years before the spread of AI, rightsholders in the EU could prohibit the use of their content for Text and Data Mining (automated analysis of text and data) through a TDM opt-out (a reservation against such use). Providers of general-purpose AI must be able to identify these reservations and comply with them.
But do they? The disputes now reaching European courts show that the practical answer is not obvious. It is impractical to read all content on the web, keep track of every restriction and preserve reliable evidence that an AI model was trained only on data it was allowed to use while the rest was actually left out.
A legal mechanism exists. Technical tools exist. But they remain difficult to use as one system. This problem led us to an idea that happened to meet a European Commission public consultation on how TDM opt-outs should work in practice.
Our vision
In January 2026, we thought that creating a single European registry of Text and Data Mining restrictions operated by EUIPO could solve two problems at once: an EU agency has the institutional budget to build a technically more advanced solution than today’s patchwork of private protocols, while the EU has the regulatory leverage to make that solution legally meaningful across the entire Single Market. The CLAIMS proposal is available here.
We proposed:
- one place for the whole EU — where a rightsholder can specify the conditions under which their work may be used for AI training;
- allowing those conditions to be stated in plain language — with the system translating them into a machine-readable format;
- giving AI companies a single API for checking — those conditions;
- linking the information to the work itself — rather than only to a particular website or file;
- allowing not only a complete opt-out — but also more specific conditions of use;
- verifying who registered the information — and recording the date on which it was entered.
We see this as a small example of Web 4.0: instead of fragmented lists and databases, there is a single verified space in which a fact about the real world is directly linked to its legal and business consequences.
What happened next
| What CLAIMS proposed | What happened in the EU | What we got |
|---|---|---|
| One place to check a TDM opt-out. | In July 2026, the European Commission published a study on the feasibility of creating an EU-wide TDM opt-out registry. | The idea of a single registry reached the stage of practical discussion at EU level. The registry itself does not yet exist. |
| Link the reservation to the work, not only to a URL. | The Commission's study specifically considers work identifiers and digital fingerprints of content. | Rights could follow the work, even if it appears on another website or in another file. |
| Do not replace existing technologies with yet another protocol — bring the information together. | The Commission concluded that a possible registry should complement existing TDM opt-out mechanisms rather than replace them. This follows from the published study of the possible registry architecture. | Instead of creating yet another separate way to express a reservation, the idea is moving towards one place where these reservations can be checked. |
| Use EUIPO as the basis for this infrastructure. | In its announcement of the study results, the Commission links further work to EUIPO's work on developing copyright infrastructure for the digital and AI era. | EUIPO has indeed become central to the further work, although the operator of any future registry has not yet been determined. |
Outcome of the initiative
As with the first initiative, we cannot claim that the European Commission did this because of CLAIMS: the idea of an EU-wide TDM opt-out registry was being studied independently of our consultation response. However, the specific model we proposed overlaps in many respects with the direction of the Commission's subsequent work.