In July, MEP Axel Voss released his draft for an own initiative report on Copyright and generative AI in the JURI Committee. While the proposal was well-intentioned, in our initial reaction we were quick to criticize numerous aspects of the text, most notably its disregard for the existing legal framework, ambiguous proposals and lack of users’ rights safeguards.
Many of our concerns were shared by MEPs most notably from the Greens and S&D who submitted their amendment proposals, aiming to shape the report in a way that strengthens creators without causing legal uncertainty. In total, members of the JURI committee tabled 370 amendments, leaving the rapporteur and the shadows with their work cut out to come up with compromise proposals to reconcile the at times widely diverging interests. To reach a consensus in the committee, the rapporteur has thus far proposed four batches of compromise amendments, covering all articles of the proposal and some of its recitals. If adopted as is, these compromises would very noticeably improve the original draft report by introducing important safeguards and addressing many of the draft’s ambiguities. However, some areas of concern remain unaddressed thus far, and worse, the latest compromise proposal contains new language which threatens to undermine the existing legal framework anew.
What has been accomplished so far?
Perhaps the most significant area of improvement within the compromise amendments (Batches 1; 2; 3) is the overall attitude towards the applicability of the TDM exceptions to the training of generative AI systems. Whereas the original draft for the own initiative report by MEP Voss stated that the training of generative AI systems was not covered by the existing framework, the current version of compromises refers to “ambiguities” in the application of the TDM exceptions to the context of training generative AI models, recommending “the swift clarification on its applicability and implementation.” The latest proposal also rather vaguely calls for an additional legal framework aimed to clarify licensing rules and address potential copyright infringements.
The current state of these articles is certainly an improvement over the original draft. In fact, in our first reaction to the draft report, we heavily criticised the claim that the TDM exceptions established in Articles 3 & 4 of the DSM Directive do not apply to the training of generative AI. Our main gripe with the approach was that this line of argumentation challenges the interpretation provided for in the AI Act and by multiple statements by the Commission and opens the door for discussions around the legality of current training practices. It should, however, be noted that previous compromise proposals had gone so far as to fully delete the sections in question. Unfortunately, the negotiation team seems to be unable to agree on a complete removal of this section from the report.
An additional area of concern that we highlighted in our review of the draft report focused around the requirements for full transparency reporting and a proposed “irrebutable” presumption of use for cases where these transparency requirements are not met. While the current compromise proposal still makes several references to “full transparency”, it overall loosens the report’s transparency requirements to align them more closely with the requirements in the AI Act and the General-Purpose AI Code of Practice. This includes an apparent shift from transparency of training data towards transparency of opt-out compliance. In addition, the presumption of use in the current text is now rebuttable.
Finally, on a positive note for users’ rights, the compromise proposals have added crucial safeguards for generating content that falls within the scope of protected uses. Specifically, the proposed text urges 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, provided that such measures do not result in the prevention of the production of GenAI outputs that include works or other subject matter that do not infringe copyright and related rights, including for private use, quotation, criticism, review, caricature, parody and pastiche, and incidental inclusion.”
There is more to be done!
The proposal also foresees certain changes when it comes to the use of press publishers’ and media content for the training of generative AI. It recommends the Commission to explore how the press publishers right and the broadcasting right can be extended to provide rightsholders from the press and media sector with “full control” over the use of their content for AI training and “for purposes beyond AI training such as inferencing and retrieval-augmented generation”, suggesting that these uses may only take place with “explicit consent,” seemingly removing this body of work from the scope of the TDM exceptions.
Finally, the latest proposal attempts to bend the concept of territoriality, suggesting that EU copyright law — in particular the rules regarding the training of generative AI systems — should apply to any AI model placed or made available on the Union market, regardless of the jurisdiction under which the training takes place. While this notion is already mentioned in the Recitals of the AI Act, an EP own initiative report is certainly not the correct place to try and introduce this interpretation.
What needs to happen next?
The most recent batches of compromise amendments we have seen for the JURI Committee’s own initiative report on copyright and generative AI are overall a step in the right direction. By addressing some of the most crucial problems with the original draft report, such as issues relating to legal uncertainty and users’ rights, the negotiation team has defused several of our initial concerns. However, more work is needed to ensure that the text is coherent with and complements the existing legal framework.
Only when the aforementioned shortcomings are addressed will the report have the opportunity to express a clear and coherent message and contribute towards setting the agenda for the upcoming Commission review of the DSM Directive.
