Changes from report parliamentary committee draft to plenary report

RemovedK. whereas this new and specific form of use (GenAI training) requires a clarification of the legal conditions under which such training may be conducted;

AddedM. whereas key legal questions about the interplay between GenAI and copyright and related rights are whether the use of copyright-protected works and other subject matter in training datasets is lawful under EU and national Member State law, what the legal status of AI-generated content should be, and how to ensure transparency, consent and the fair remuneration of creators and rights holders when their protected works and other subject matter are used in the generation, dissemination and distribution of AI outputs;

RemovedL. whereas high-quality and comprehensive training datasets are essential for the effective development of GenAI systems and to secure high-quality and trustworthy outputs of GenAI systems; whereas enabling the lawful use of such datasets within the European Union is therefore crucial to fostering innovation, ensuring technological sovereignty, and maintaining the Union’s competitiveness in the rapidly evolving global AI landscape;

AddedN. whereas GenAI, by massively and inexpensively creating content that mimics what was produced by human creativity, competes directly with the work of creators, particularly of cultural and media content; whereas this competition may lead to a decline in the quality of online content as a result of creators disengaging, and thus to an overall decline in human cultural and creative production;

RemovedM. whereas the upcoming launch of the EUIPO Copyright Knowledge Centre would represent a timely and commendable initiative aimed at strengthening the interface between copyright and emerging technologies, particularly GenAI;

AddedO. whereas this risk of a gradual disappearance of the human dimension of creation in favour of AI-generated content, in addition to undermining the economic viability of the creative sector, poses an existential risk to European society and democracy, in that it blurs the boundaries between truth and falsehood, the perception of discourse and its authors, and disrupts cognitive faculties and critical thinking; whereas the Artificial Intelligence Act (AI Act) takes these risks into account, but cannot be complete without robust protection of copyright and related rights;

RemovedN. whereas, in addition to a standardised machine-readable opt-out, rights holders should also have the possibility to register such opt-out in a centralised registry, using a single technological standard and in machine-readable format, potentially managed by the European Union Intellectual Property Office (EUIPO), thereby enabling the effective exclusion of registered works from automated data crawling;

AddedP. whereas to ensure proper enforcement of the law and a level playing field across the EU and among AI providers, the EU’s and Member States’ rules on copyright and related rights need to apply uniformly to all AI providers deploying products or offering services within the EU, irrespective of their place of establishment, of the jurisdiction in which the copyright-relevant acts underpinning the training of those AI models take place, and of where the output produced by the AI system is used in the EU; whereas the same requirement should apply mutatis mutandis to any subsequent use of content for inference, retrieval-augmented generation or fine-tuning, not only by providers of AI models, as currently stipulated by Article 53 of the AI Act, but also by providers or deployers of AI systems; whereas entering into new licence agreements should not be misconstrued as redress for past unauthorised uses of copyright-protected content;

RemovedO. whereas any GenAI provider should ensure full and detailed transparency concerning all copyright-protected content used to train that system, irrespective of the jurisdiction in which the copyright-relevant acts underlying the training were performed; whereas this transparency shall consist in an itemised list identifying each copyright-protected content used for training; whereas the same requirement should apply mutatis mutandis to any subsequent use of content for inference, retrieval-augmented generation or fine-tuning not only by providers of AI models, as currently stipulated by Article 53 AIA, but also by providers or deployers of AI systems;

AddedQ. whereas any AI provider placing a general-purpose AI model on the EU market must remain responsible for verifying that the measures included in their copyright policy, as outlined in the General-Purpose AI Code of Practice, comply with Member States’ implementation of EU law on copyright and related rights, before carrying out any copyright-relevant act in the territory of the relevant Member State, as failure to do so may give rise to liability under EU law on copyright and related rights;

RemovedP. whereas such transparency could be facilitated through a trusted intermediary, such as the European Union Intellectual Property Office (EUIPO), which would be responsible for notifying rights holders of the use of their content, thereby enabling them to assert claims in relation to its use for training; such an intermediary should be endowed with the necessary powers and resources to assess whether providers and deployers comply fully with the transparency obligations;

AddedR. whereas the Code of Practice is only of voluntary compliance, and not all AI providers have signed it, in particular its copyright chapter;

RemovedQ. whereas, as an alternative to the aforementioned EUIPO register, transparency could also be achieved by enabling rights holders to watermark their works and other protected subject matter, and by requiring AI providers to make available search tools that allow for the detection of such watermarks among the materials used for training;

AddedS. whereas the global nature and exponential rate of AI training, deployment and offers on the one hand, and the territorial application of the rules pertaining to copyright and related rights on the other hand, create obstacles for obtaining licensing on the relevant rights and for making AI products and services available on the EU market compliant with those rules and rights, in a way similar to but dramatically increased to the digital non-AI era;

RemovedR. whereas, in addition to the obligation of full transparency concerning copyright-protected works and other protected subject matter, there is a need to establish a mechanism whereby, under certain conditions, the failure by AI providers or deployers to provide complete transparency shall give rise to an irrebuttable presumption that any relevant copyrighted work or other protected subject matter has been used for training purposes, thereby triggering all applicable legal consequences under Union and national law for the infringement of copyright or related rights; whereas, where a court finds in favour of a rights holder on the basis of either such a presumption or of submitted evidence, all reasonable and proportionate legal costs and other expenses shall be borne by the AI provider;

AddedT. whereas case-law on the topic of the infringing use of copyrighted content for training AI models is still scarce, but several court decisions are expected in the near future, since judicial actions have been brought by various rights holders in the EU and other parts of the world against certain AI providers;

RemovedS. whereas there is a knowledge gap among GenAI providers, especially smaller firms, regarding their copyright obligations under EU law;

AddedU. whereas GenAI training, a specific form of use, requires clarification of the legal conditions under which such training may be conducted;

RemovedT. whereas the press sector holds a vital role in safeguarding democracy and the democratic structure within the European Union; whereas it is essential to ensure that GenAI models and systems do not engage in selective processing that favours certain publications over others, thereby preserving the plurality and impartiality of information; whereas GenAI models and systems must be designed to incorporate and consider the full spectrum of press publications to uphold fundamental democratic values of diversity and fairness in public discourse; whereas there is a need to establish clear quality standards for GenAI models and systems;

AddedV. whereas high-quality, human-based (non-synthetic) and comprehensive training datasets are essential both for research and for the effective commercial development of GenAI systems and for securing high-quality and trustworthy outputs from GenAI systems; whereas enabling the lawful creation and use of such datasets within the EU, which excludes pirated, unauthorised or otherwise infringing material, is therefore crucial to fostering innovation, ensuring technological and cultural sovereignty, maintaining the EU’s competitiveness in the rapidly evolving global AI landscape and protecting Europe’s cultural and creative industries;

RemovedGenAI output

AddedW. whereas the launch of the European Union Intellectual Property Office (EUIPO) Copyright Knowledge Centre would represent a timely and commendable initiative aimed at providing useful and reliable information on copyright and strengthening the interface between owners of copyright and related rights and representatives and stakeholders acting in the field of emerging technologies, particularly GenAI;

RemovedU. whereas transparency regarding the output generated by artificial intelligence systems is essential to enable proper classification of works as ‘human-created’ or otherwise; whereas such classification entails significant legal consequences, including for the applicability of copyright protection and the determination of rights and liabilities;

AddedX. whereas rights holders should have the possibility to use tools enabling them to effectively exclude the use of their works in AI training in a limited number of standardised machine-readable formats, potentially managed and listed by the EUIPO, thereby enabling the effective exclusion of registered works from automated data crawling and providing legal certainty to both rights holders and AI providers;

RemovedV. whereas the generation of outputs, characteristic of GenAI, can infringe the rights of reproduction, of making available to the public, or of communication to the public; whereas the TDM exception as provided for in article 4 of the CSDSM does not cover the right of making available to the public or the right of communication to the public;

AddedY. whereas any provider or deployer of GenAI models and systems should ensure transparency concerning all copyright-protected content used to train those models and systems, irrespective of the jurisdiction in which the copyright-relevant acts underpinning the training of their GenAI models take place; whereas this transparency must consist of an itemised list identifying each item of copyright-protected content used for training; whereas the same requirement should apply mutatis mutandis to any subsequent use of this content for other purposes, including inference, retrieval-augmented generation or fine-tuning, not only by providers of AI models, as currently stipulated by Article 53 of the AI Act, but also by providers or deployers of AI systems; whereas uses such as inference and retrieval-augmented generation essentially happen continuously and in real-time; whereas, in such cases, transparency should include the act of crawling itself, whereby, on the one hand, crawlers are required to identify themselves to web operators, and, on the other hand, AI providers or deployers are required to maintain detailed records of their crawling activities; whereas a presumption that content has been crawled, including for inference and retrieval-augmented generation purposes, should apply where AI systems handle user queries; whereas the mere information of the third-party content used by AI and GenAI providers and deployers does not constitute trade secrets under EU law;

RemovedX. whereas, when it comes to the legal treatment of GenAI outputs, EU copyright law remains grounded in the principles of human authorship; whereas according to the settled case law of the Court of Justice of the European Union, the concept of a ‘work’ entails two cumulative conditions: first, it must be an original subject matter that reflects the author’s own intellectual creation; second, that creation must be expressed in a manner that makes it identifiable with sufficient precision and objectivity;

AddedZ. whereas such transparency could be facilitated through a trusted intermediary, such as the EUIPO, which would be responsible for notifying rights holders of the use of their content, thereby enabling them to assert claims in relation to its use for training; whereas such an intermediary should be endowed with the necessary powers and resources to assess whether providers and deployers comply fully with their transparency obligations;

RemovedY. whereas inconsistent international regulation regarding the copyright eligibility of AI-generated content poses a risk to the global coherence of intellectual property law and may give rise to regulatory arbitrage or undermine the competitiveness of the Union’s creative and AI sectors; whereas international convergence and the establishment of a global regulatory framework would provide a more effective and coherent alternative to the current fragmentation of legal approaches;

AddedAA. whereas, transparency could also be achieved by enabling rights holders to watermark their works and other protected subject matter, and by requiring AI providers to keep such watermarking unaltered, and to make search tools available that allow for the detection of such watermarks in materials used for AI training;

RemovedZ. whereas, to the extent permitted by international law, the existing principle of territoriality needs to be adapted for the training of GenAI systems in order to ensure that training with European content is subject to European law even if it is realised outside the EU;

AddedAB. whereas, in addition to the obligation of transparency concerning copyright-protected works and other protected subject matter, there is a need to establish a mechanism whereby, under certain conditions, the failure by AI providers or deployers of AI models and systems to provide complete transparency should give rise to a rebuttable presumption that any relevant copyright-protected work or other protected subject matter has been used for the purposes of training, inferencing or retrieval-augmented generation, thereby triggering all applicable legal consequences under EU and national law for the infringement of copyright or related rights; whereas, where a court finds in favour of a rights holder or the organisations representing them on the basis of either such a presumption or of submitted evidence, all reasonable and proportionate legal costs and other expenses should be borne by the AI provider;

Removed1. Recommends that the Commission, independently of its planned review of the copyright framework and the CDSM Directive and without presupposing the need for legislative revision, urgently conduct a thorough assessment of whether the existing EU copyright acquis adequately addresses the legal uncertainty and competitive effects associated with the use of protected works and other subject matter for the training of generative AI systems, as well as the dissemination of AI-generated content that may substitute human-created expression;

AddedAC. whereas the press sector plays a vital role in safeguarding democracy and democratic structures within the EU; whereas it is essential to ensure that GenAI models and systems do not engage in selective processing that favours certain publications over others, thereby preserving the plurality and impartiality of information; whereas GenAI models and systems must be designed to incorporate and consider the full spectrum of press publications in order to uphold the fundamental democratic values of diversity and fairness in public discourse; whereas there is a need to establish clear quality standards for GenAI models and systems;

Removed2. Further recommends that such assessment aims to uphold a framework in which fair remuneration mechanisms enable the generation of the resources needed for European artistic and creative production to thrive in the context of AI-driven global transformation;

AddedAD. whereas transparency regarding the output generated by AI systems is essential to enable proper classification of content as ‘AI-generated’ or ‘human-created’, depending on whether the output meets the established criteria for copyright protection; whereas such classification entails significant legal consequences, including for the applicability of copyright protection and the determination of rights and liabilities; whereas citizens’ trust in AI can only be built on a regulatory framework, which ensures that any AI system put into operation fully respects and complies with the Treaties, the Charter and secondary EU law;

Removed3. Notes the use of generative AI systems that rely on protected content without authorisation from, or compensation to, the rights holders affected by such use, particularly when integrated into search engines or other digital services that enable the generation, often in real time and at marginal cost, of content that imitates or directly draws upon original works and other protected subject matter on which the models were trained or that was scraped, including in real time, by such models; is alarmed that these practices may result in the provision of products and services that directly compete with those of the rights holders;

AddedAE. whereas, in terms of the legal treatment of GenAI outputs, EU copyright law remains grounded in the principles of human authorship; whereas according to the settled case-law of the Court of Justice of the European Union, the concept of a ‘work’ entails two cumulative conditions: first, it must be an original subject matter that reflects the author’s own intellectual creation, and second, that creation must be expressed in a manner that makes it identifiable with sufficient precision and objectivity;

Removed4. Calls on the Commission to immediately impose a remuneration obligation on providers of general-purpose AI models and systems in respect of the novel use of content protected by copyright or related rights, with such obligation applying until the reforms envisaged in this report are enacted;

AddedAF. whereas GenAI is increasingly used as a tool to aid the enforcement of copyright and related rights; whereas the Commission should ensure that the use of GenAI for enforcing copyright claims is subject to effective safeguards, including safeguards against misuse, ensuring full respect for fundamental rights, especially freedom of expression and freedom of the press;

Removed5. Encourages the Commission to coordinate efforts on raising awareness on copyright among AI developers, which may include compliance checklists, legal and technological toolkits, and technical guides;

AddedAG. whereas inconsistent international rules on the eligibility of AI-generated content for copyright protection poses a risk to the global coherence of intellectual property law and may give rise to regulatory arbitrage or undermine the competitiveness of the EU’s creative and AI sectors; whereas international convergence and the establishment of a global regulatory framework would provide a more effective and coherent alternative to the current fragmentation of legal approaches; whereas EU regulation on AI and copyright should not be impacted by external political pressure; whereas, in this context, the EU must offer a strong and united response to secure the functioning of its internal market, protect its workforce and competitiveness, strengthen its technological autonomy and set global standards in the field of intellectual property;

Removed6. Supports the clarification of the TDM exception under Article 4 CDSM as regards the main flaws and ambiguities detected thus far in its application, especially as concerns the establishment of a clear machine-readable standard for the opt-out and the concept of ‘lawful access’;

AddedAH. whereas, in order to ensure the proper implementation of EU law in relation to all generative AI services and to prevent non-EU-based providers from gaining an unfair competitive advantage through non-compliance, it is essential that rules protecting and enforcing authors’ rights at EU and national level apply effectively to all services operating on the EU market; whereas the growing concentration of power in the hands of a few large non-European companies increases the EU’s strategic dependence and weakens the capacity of European creators and the European press to exercise and enforce their rights; whereas, to the extent permitted under EU and international law, the principle of the territoriality of copyright protection must be adapted to the training of generative AI systems so as to ensure that the use of European content is subject to EU law even when such training takes place outside the EU, which is fundamental to guaranteeing fair remuneration for European creators and rights holders and to safeguarding the cultural and creative vitality of the EU, as well as to ensuring fair competition between European and non-European providers of GenAI systems; whereas GenAI systems that do not comply with these requirements should be barred from operation within the EU and these principles should be robustly enforced;

Removed7. Believes that a legal framework for GenAI should be established either through the introduction of a dedicated exception to the exclusive rights to reproduction and extraction, distinct from that provided for TDM under Article 4 of the CDSM Directive, or by expanding the scope of that provision to explicitly encompass the training of GenAI, which is currently not covered; stresses that rights holders shall have the right to opt out through a standardised, machine-readable mechanism;

AddedAI. whereas the rapid advancement of GenAI has significantly increased the possibilities to create and disseminate realistic manipulated digital images, audio or video content, including artists’ works and performances, that resembles or imitates real people, known as ‘deep fakes’, which falsely appear to be authentic or truthful; whereas the misuse of such content poses a risk to these people’s identity and their personality, including their body, facial features and voice;

Removed8. Recommends that the Commission ensures the compatibility of this new GenAI legal framework with the three-step test of Article 5(5) InfoSoc Directive;

Added1. Notes the ambiguities thus far in the application of the CDSM Directive in the context of GenAI training and recommends the swift clarification on its application and implementation;

Removed9. Recommends assigning the EUIPO responsibility for setting up and managing a central register of opt-outs and, where necessary, for mediating the licensing process, so as to streamline relations between GenAI providers and rights holders, establishing a workable, innovation-friendly framework that supports the Union’s competitiveness without unduly hindering the development of AI technologies; further recommends that both opt-out declarations and licence offers be recorded in machine-readable form in the same register;

Added2. Believes that current copyright law is insufficient to address the challenge of licensing copyrighted material for GenAI; calls for an additional legal framework to clarify licensing rules for GenAI and to address potential infringements of current copyright law; insists that such a framework should include provisions ensuring the effective cooperation of GenAI providers with creators and other rights holders, including a functioning licensing market that restores the bargaining power of rights holders and viable protection solutions;

Removed10. Calls on the Commission to propose the full, actionable transparency and source documentation by providers and deployers of general-purpose AI models and systems, with regard to the use of any copyright-protected work or other protected subject matter for any purpose, including for inferencing, retrieval-augmented generation, or fine-tuning, taking into due account the need to protect trade secrets and confidential business information;

Added3. Recalls that any exceptions to current rights under the copyright acquis, including on AI, must be in line with the three-step test under Article 5(5) of the InfoSoc Directive and under the Berne Convention; notes that the concept of ‘lawful access’ excludes pirated and infringing copies of works;

Removed11. Calls on the Commission to propose the establishment of an irrebuttable presumption that, for any general-purpose AI (GenAI) model or system placed on the Union market, works and other subject matter protected by copyright or related rights have been used for its training where the statutory transparency obligations set out in this resolution have not been fully complied with; further recommends that, where a rights holder succeeds in legal proceedings either on the basis of this presumption or through submitted evidence, any reasonable and proportionate legal costs and other expenses incurred in enforcing such rights shall be borne by the provider of the AI model or system;

Added4. Calls on the Commission to ensure that activities conducted for scientific research or educational purposes, in particular by research organisations and cultural heritage institutions, or in the framework of non-commercial innovation, are not restricted, in line with the principle contained in Article 13 of the Charter; urges the Commission to ensure that the commercial exploitation of research results stemming from these activities is not restricted, provided that adequate authorisations have been obtained from rights holders;

Removed12. Insists that AI-generated content should remain ineligible for copyright protection, and that the public domain status of such works be clearly determined;

Added5. Reiterates that rights holders, especially from the press and news media sector, and especially press publishers, journalists and news editors, must have full control over the digital use of their content by AI systems and models for training purposes; stresses that this control should be based on a robust and functioning possibility to exclude such use by AI systems and models, underpinned by full transparency and source documentation over the use of third-party content;

Removed13. Calls on the Commission to explore measures to counter the infringement of the rights of reproduction, of making available to the public and of communication to the public through the production of GenAI outputs;

Added6. Believes, furthermore, that said rights holders must have full control over the use of their content for purposes beyond AI training, such as inferencing and retrieval-augmented generation by systems such as applications, or the use for purposes conducive to the production of AI-generated, competing offerings in the primary markets of the rights holders; believes that the use of protected content for such purposes beyond training can take place only with the express consent of the rights holders; suggests that the Commission explore how ancillary rights for press publishers, journalists and news editors, and other related rights, including for news media producers and news broadcasters, could be extended to cover those purposes; believes that such remuneration rights may also be managed under a voluntary collective licensing system, whereby a presumption of collective rights management on behalf of rights holders with respect to Al-related uses of their content could be envisaged, without prejudice to the right of rights holders to opt for individual licensing;

Added7. Urges the Commission to explore the possibility to safeguard the press and news media sector, whose services are repeatedly and fully exploited by AI systems, and to assess and, where appropriate, propose mechanisms ensuring that providers of GenAI models or systems that demonstrably divert traffic and revenue from press and news media outlets compensate such outlets in a fair, proportionate and non-discriminatory manner, paying particular attention to local and regional media, in order to safeguard media pluralism, diversity and democratic discourse in the EU and its Member States;

Added8. Notes the growing use of GenAI to aggregate or present news content and its far-reaching effects on access to and diversity of information; calls on the Commission and the Member States to ensure the full enforcement of EU law to address this phenomenon; stresses that such systems should uphold and contribute to the principles underpinning the European Media Freedom Act, in particular media pluralism and diversity of information; underlines the need for proper enforcement of Article 34 of the Digital Services Act when such providers are designated as very large online platforms or search engines, ensuring that systemic risks to media freedom and pluralism are properly addressed; calls further on the Commission to swiftly address any violations of the Digital Markets Act with regard to potential self-preferencing practices by gatekeepers benefiting their AI services and harming fair competition with business users;