Last week, the Commission sought input from rightsholders and GPAI model providers to find out if 55 months after the entering into force of the so-called TDM-exception, they had already been able to put in place protocols that would allow right holders to reserve their rights in a machine-readable manner. A 40-page long survey will most probably provide the Commission with lots of insights. However, the most important question was missing: how can rightsholders and AI developers work together to develop a protocol that is accessible by performers and authors who do not have access to the location nor asset via which their work is made accessible for crawlers?
The survey is linked to the Copyright chapter of the General-Purpose AI Code of Practice, which was finalised in 2025. In measure 1.3. of that Code, AI providers commit to employ web-crawlers that read and follow instructions expressed in accordance with the Robot Exclusion Protocol (robots.txt) as well as to identify and comply with any other appropriate machine-readable protocol that has been adopted by international or European standardisation organisations or is “widely adopted by rightsholders”.
The Commission now seeks views on that second category and is inspired by a recent study of the EUIPO in which seven different technical solutions were identified as having potential of being widely adopted. Its purpose is to find out which of these can be retained in a published list of generally agreed machine-readable opt-out solutions. Such solutions would no longer need to be identified, and compliance would apply automatically for the signatories of the Code.
At this stage it is important to not limit further the already limited obligations that come with the signing of the Code of Practice. It is important to create room for optimal nuanced sector-specific solutions to develop and facilitate the compliance of signatories with their obligation to respect all protocols used to express a rights reservation.
So, one can look at this as a perfect step by the Commission to help rightsholders find a way out of the misery that Article 4 has brought about. Especially when a revision of that article and the exception contained therein appears to be increasingly out of the question.
However, with this approach the Commission is missing a major opportunity to arrive at a workable solution. Rather than focusing on the technical feasibility of these protocols, attention should be paid to accessibility for non-owners of assets and locations. Especially with the ever-increasing convergence within the media landscape, the Commission must ensure that the right to make a reservation of rights enforceable does not become the privilege of a limited number of major players.
Individual authors and performers have the right to not authorise the use of their work for AI purposes by means of a rights reservation. But they face a very practical impracticality. Once their performance has been incorporated in a phonogram or videogram, they have no control of whether any metadata or other means of identification will be attached.
In particular, they have no control over whether metadata will be included that would indicate a performer’s preference when it came to rights reservation. The problem is not limited to performers and authors. Producers of different types of cultural content face the problem of not having control over the asset or location of their productions. This impracticality persists even in the situation where these rightsholders decide to entrust their collective management organisations with managing their choice for rights reservation.
For this particular reason, any protocol put forward should be open in a non-discriminatory way to all rightsholders in the sector. In the audio and audiovisual sectors, performers, authors and producers should all be able to engage with a protocol also in situations where they have no access to the asset or location via which their work has been made accessible for TDM activities. Such access can be provided by the establishment of a centralised registry of rights reservations. The work that the Commission is doing on this matter should therefore also focus on the problems of inaccessibility which the vast majority of rightsholders face.
It cannot be emphasized enough that the Commission’s approach here is in stark contrast to the way it is guiding AI providers to develop a standardised way to label their synthetically generated content in a machine-readable manner. Like a helicoptering parent it is doing everything in its power to help AI developers with yet another Code of Practice, while actively pushing member states to postpone the entering into force of the underlying obligations under the AI Act. A similar approach where rightsholders were brought together while the entering into force of the TDM exception was postponed was not given to the creative sector after the adoption of the 2019 CDSM directive.
In the most cynical way, AI providers oppose the one-size-fits-all approach and the use of robots.txt which they considered to be unfit for purpose. They criticise the draft of the Code of Practice on Transparency of AI-Generated Content for going beyond current capabilities of most AI companies. They fear having to retrofit complex watermarking and request the Code starts by setting out realistic baselines. For several years now, rightsholders have expressed similar criticism towards the impossibility of making their right reservation under the TDM exception practically possible.
And so we agree with the additional finding of the EUIPO study that there is a “need for AI developers to engage more actively with the creative and cultural sectors on effective ways to account for the TDM opt-out expressed” and are confident that if this happens, common solutions can be found.
Performers and their CMOs are eager to explore and develop any technical reservation measures which would result in an efficient rights reservation solution for all.
