Changes from report parliamentary committee draft to plenary report

With the ever-advancing digital and technological development, notably as regards AI and the so-called Generative Large Language Models, some rights such as copyright, the right to privacy and the right to non-discrimination are being devalued as it is becoming extremely difficult to enforce them without an enormous legal and financial risk to the rights holders.

This creates huge legal uncertainties for all parties involved. However, if the European legislator, through its lengthy procedures and a lack of courage, continues to refuse to tackle the crucial issues head on, the EU and its actors will always be left at a disadvantage and further dependencies will be created.

Therefore, the principle that must apply is that technological developments must respect existing laws while, on the other hand, existing laws must not hinder technological developments. This urgently calls for workable solutions, which are currently not being provided by market participants on either side, with technological developments colliding with copyright, such that they appear no longer compatible with each other. It is likely that perfect, comprehensive solutions will no longer be possible.

For this reason, the European legislator needs to strike a fair balance between the interests of all stakeholders at the earliest opportunity. It would also be desirable to find a permanent workable solution to avoid having to protect copyright claims against new technologies every five or six years. A ‘General Copyright Protection Regulation’, akin to the General Data Protection Regulation, could be helpful in this regard. The AI liability proposal ought to have provided procedurally swifter solutions. However, owing to a lack of strategic long-term vision on the part of various actors, this is now to be withdrawn (this being probably legally dubious) so that legal uncertainty will persist.

This report therefore is an attempt to bring a workable balance between new technology and copyright closer. This will require a combination of legal, technical and technological solutions.

The political context for the EU is not easy because:

1) in geopolitical terms, guiding values are diverging in the ‘West’;

2) European AI development is severely lagging behind and needs to be promoted without additional ‘obstacles’;

3) the current added value in digital development is being generated by large tech companies in the US, to the detriment of the European creative sector.

This report therefore also aims to support and promote the development of AI in Europe. After all, Europe needs AI to drive the digitalisation that is essential in our globalised world. The opportunities for European progress are immense and must under no circumstances be left untapped. However, we also want AI systems in Europe that meet certain requirements as regards quality and trustworthiness, and this can only be achieved with quality data belonging to content creators.

At the same time, we also want to preserve copyright protection for the works of our creative sectors and cultural professionals. European culture constitutes a fundamental part of our identity. European added value in the creative sectors is immense and it should not be possible to use it without compensation. This novel use of protected content (training data, data used for generative output) must therefore be remunerated. Only thus can European cultural professionals create the economic basis for the generation of further content.

In this context, the copyright-related rights relating to the content of press publishers play an additional special role for our European understanding of democracy and the rule of law. Freedom of the press, freedom of opinion and freedom of information must under no circumstances be undermined or subjugated by artificial intelligence, in particular where, as is currently the case, digital access and digital ‘distribution’ of information are increasingly taking place through search engines and AI and, in the case of generative AI, are also influenced by bots, all this being in the hands of a few companies. The control of information and disinformation, whether bot-driven or not, as well as deepfakes and the resulting interference in elections or influencing of public opinion, is extremely dangerous. Plurality and diversification of opinion in this context must therefore be guaranteed in the form of an independent press. This means that the processing of content in an automated and generative manner must trigger a commensurate compensation. However, it is also important for press publishers that they remain identifiable to users. This may require an obligation to cite sources.

In this connection, the comparable situation of so-called paywalls, the data behind which is not always solely copyright-protected material – though at times it is – should also be examined.

However, under the current situation, it must also be assumed that a slowly developing licensing market will no longer encompass all market participants. This means that not every press publisher or not all copyright-protected content is needed to provide training material for a fee, so that market participants lacking in bargaining power may no longer be taken into consideration if they demand compensation. A solution, if it is even necessary at this point, will probably only be possible by means of flat-rate fees or a stronger organisation to pool bargaining power. It is, however, questionable whether all parties would want to embark on such a path.

While this should mainly be left to the market participants, the question arises – given the generative AI systems that are offered worldwide –as to whether only ‘global licences’ will in future play a role in licensing.

Since this ‘copyright-protected’ data has been used for years (at least since late 2022) without a licence or other authorisation from the creators, consideration must also be given to the extent to which compensation should also be paid retroactively.

In any event, the European legislator or the European Commission should, pending the introduction of an appropriate provision to address this problem, establish an immediate, simple, flat-rate copyright fee for this use of 5 to 7% of global turnover in order to compensate for the added value that these businesses generate using the data of European creatives and to ensure it remains in Europe.

Of course, this also calls for a reassessment of the territoriality principle, as already envisaged in the AI Act. We cannot allow AI models to be trained just anywhere in the world using European copyright-protected data only for them to be then made available in Europe.

In the future, there may also be a need for a democratic legislator having to either review or standardise the quality of the basic data used by generative ‘Large Language Models’. In the future, it may also be necessary for the legislator to require such AI developers to include copyright-protected works in order to maintain the quality of those models.

On the other hand, we should also use AI to enhance the independence and diversity of high-quality information.

AI also needs access to copyright-protected works in order to develop further in terms of quality. In recent years, therefore, a huge amount of content has been used by AI developers mainly, but not solely, for training purposes. Such training constitutes a completely novel use of content, to which the existing copyright rules are applicable only to a limited extent.

Currently, the lack of compensation results in enormous legal uncertainty. A legally clear solution to this source of conflict is required with the utmost urgency. As long as there is no clear legal framework regulating conflicting interests, European content will be used with the added value being generated elsewhere. The Commission must therefore take immediate action without waiting for possible reviews of, for instance, the Copyright Directive or the AI Act.

A solution must arise from the interplay between licensing possibilities and transparency requirements on the basis of international law and the resulting right of disposal of the content creator.

Thus, a reference to Article 4 of the DSM Directive is still not sufficient and probably not in line with international copyright principles. The European legislator incorporated Article 4 in the AI Act without, however, having clearly established the consequences. The current exemption in Article 4 allowing text and data mining under the conditions specified was not drafted with the intention of enabling the use en masse of copyright-protected material by all through generative AI. And certainly not where it also leads to the creation of a competitive product accessible to the public.

Nevertheless, even if in the opinion of the rapporteur Article 4 is not applicable, we need a similar possibility allowing developers of AI to obtain licences for copyright-protected works in as straightforward and technically easy to implement manner as possible. This is best done digitally. Moreover, the rationale of Article 17 of the DSM Directive should also be taken into account.

At the same time, it is important to ensure that rights holders are still able to decide whether – or not – and how (licencing) their content can be utilised for this new type of use. Right-holders must be given the right to an opt-out.

However, for this to be recognised easily and unhindered by AI developers, it must be machine-readable and standardised. It must also be the responsibility of the rights holder to make use of this opt-out in a legally and practically certain manner.

In order to make implementation as simple as possible for AI developers, it seems necessary to record the opt-out in a European register. It would therefore seem appropriate for such a register to be maintained by EUIPO. The AI developer would thus have the possibility either to respect the standardised, machine-readable opt-outs or to identify through the register which works may not be used without permission.

On the other hand, only content protected with an opt-out or registered can still be protected from this type of use. It would even be possible to link the licensing process to this register, thereby simplifying matters and creating a kind of ‘one-stop-shop’ for AI developers.

As rights holders are not and cannot be aware whether their content is being used in this novel way, a transparency requirement is essential. This transparency requirement would oblige AI developers to provide a comprehensive and detailed list of the protected content they have drawn on for this novel use. The ‘sufficiently detailed summary’ provided for by the AI Act has so far been completely inadequate since it cannot provide clarity regarding the use of content precisely because it is a summary. An interpretation which can also be used in the context of copyright law is therefore necessary here.

If, for example, reasons such as trade secrets preclude access to this data base in the form of transparency that is relevant to the content creator, the obligation must be fulfilled through a trust as an intermediary. Here too, EUIPO could act as the intermediary. It could then inform the rights holders of the use made of their work.

The transparency obligation can also be fulfilled by requiring labelling of the copyrighted work, e.g. watermarks or the like, and allowing rights holders to cross-check this either through access to the basic AI model or via the register. In a digital world, a digital fingerprint on the protected works seems essential anyway.

The legislator will probably also have to resolve the issue that not every private website containing a copyright-protected image is automatically excluded from the training data.

Abuse, manipulation of information, legal assumptions, reversal of the burden of proof or even the very strong legal remedy of liability must be considered or weighed very carefully at all times and at every step of the way.

In addition to considering how to solve this problem, copyright law needs generally to be adapted to technological developments. This will require further European harmonisation of Member States’ national copyright laws.