Stakeholder consultation on AI and copyright compliance

The stakeholder consultation aims to gather written feedback from various stakeholders on the use and uptake of different technical solutions used by rightsholders to reserve their rights from the text and data mining exception (TDM) pursuant to the DSM Copyright Directive (EU) 2019/790. This stakeholder consultation is a first step in the process aimed at identifying generally-agreed state of the art TDM opt-out protocols as a follow-up to the general purpose AI (GPAI) Code of Practice measure to identify and comply with such protocols in the context of AI training (Measure 1.3 of the Copyright section of the Code) and more generally with the AI Act’s copyright-related obligation for providers of GPAI models under Article 53(1)(c) AI Act.

TDM is an automated analytical technique that analyses text and data in digital form to generate information, including patterns, trends and correlations. The DSM Copyright Directive (EU) 2019/790 introduced an exception to copyright which allows users, under certain conditions, to reproduce or extract content for TDM purposes without needing to seek the rightsholders’ authorisation. The exception also foresees that rightsholders may reserve their rights, or ‘opt out’ from the exception, which must be done in a machine-readable form for content made publicly available online. Since the development of AI models requires the use of a large amount of content, which is often extracted from the web, this exception is relevant in the context of AI training. In this respect, Article 53 AI Act obliges GPAI model providers to put in place a policy to comply with EU law on copyright and related rights, and in particular to identify and comply, including through state-of-the-art technologies, reservations of rights made under the TDM exception.    

The list of appropriate, generally-agreed machine-readable opt-out solutions supports the effective implementation of the AI Act’s obligation and Measure 1.3 of the GPAI Code of Practice. It will help GPAI Code signatories and other providers of GPAI models to demonstrate compliance with the respective AI Act obligation to identify and comply with reservation of rights expressed by rightsholders under the TDM exception.

The Copyright Chapter of the GPAI Code includes Measure 1.3 (Identify and comply with rights reservations when crawling the World Wide Web). In Point 1(b) of this measure, Signatories of the Code commit, on top of respecting robots.txt and any subsequent version of that (commitment under Point 1(a)), to identify and comply with other appropriate machine-readable protocols used to express rights reservations pursuant to Article 4(3) of Directive (EU) 2019/790 (DSM Directive). Such other protocols can either be adopted by international or European standardisation organisations or, as a more immediate solution, through a process to be facilitated at EU level. The stakeholders’ consultation is a first step in this process at EU-level.

In particular, Measure 1.3(b). of the Copyright Section of the Code requires that the opt-out protocols must be state-of-the-art, including technically implementable, and widely adopted by rightsholders, considering different cultural sectors, and generally agreed through an inclusive process based on bona fide discussions to be facilitated at EU level with the involvement of rightsholders, AI providers and other relevant stakeholders as a more immediate solution, while anticipating the development of standards.

The consultation is based on the technical solutions that have been identified in the recent EUIPO study on development of generative artificial intelligence from a copyright perspective1. In addition to robots.txt, the study identifies seven other technical solutions (TDM reservation protocol (TDMRep); C2PA TDM Assertions; AI.txt; Do not train registry (Spawning AI); JPEG Trust core foundation v2; TDM.ai protocol (Liccium); Open Rights Data Exchange (Valunode).

Rightsholders are free to use any protocols or methods to express the opt-out as long as they are fulfilling the criteria established in the DSM Copyright Directive (EU) 2019/790, including being expressed in machine-readable means in the case of content made publicly available online. The outcome of this process will be a list of machine-readable opt-out solutions generally agreed upon between signatories, rightsholders and other relevant stakeholders and will be used by the AI Office to assess compliance by signatories of the Code and other GPAI providers with their obligation under Article 53(1)(c) AI Act. This process does not affect the application of the DSM Directive or of rules pertaining to the enforcement of copyright.

The input from the stakeholders’ consultation will be used to prepare next steps of the process, i.e. online workshops with the different stakeholder groups facilitated by the Commission with a view to reaching a general agreement on a list of opt-out protocols. The consultation allows interested stakeholders to express their interest in participating to these workshops. Signatories to the GPAI Code of Practice will be invited by default. At the end of the process, a list of the generally agreed-upon TDM opt-out solutions will be shared with all participants and subsequently published on the Commission and/or EUIPO website.