Free knowledge for research

Read the report as PDF (in Swedish).

Free knowledge for research. Open norms, science and AI

Introduction

Wikipedia and the other Wikimedia platforms have evolved into some of the world's most widely used sources of free knowledge. Free knowledge is a human right and is central to equal access to information, good research, accessible cultural heritage and lifelong learning for democratic citizens.

Our free flow of knowledge has worked because legislation has allowed platforms to operate effectively and has not restricted the ability of volunteers and organisations to contribute knowledge. However, legislation is constantly evolving and we want to contribute with our unique perspective. In the three reports you hold in your hand, we holistically shed light on various fateful issues that could determine how platforms evolve in the 2020s. I hope that these insights will help us work together to develop Swedish legislation so that the soil for free knowledge is as fertile as possible.

John Andersson

Executive Director, Wikimedia Sverige

Open norms, the three-step test and freedom of panorama

We live in an era of rapid technological change. In many ways, this is a good thing. Digital platforms and technologies provide new opportunities to access knowledge and information, and the availability of knowledge favours innovation, new thinking and creative solutions.

However, in many cases it has become clear that legislation is not keeping up, with copyright being one that immediately springs to mind. Authors find their remuneration shrinking while the public is tied to subscription services locked behind paywalls and technological protection measures. The big winners are rarely those who create the works, but rather those who have acquired the rights to them. Politicians and lawyers struggle to keep up with the rapid pace of change, and their efforts often result in what amounts to a patchwork of specialised solutions and stopgap measures. Neither the public nor the creators are satisfied.

The digital world's ability to share knowledge is one of the most democratising developments in modern times. It also opens up entirely new opportunities for research and education, and digital technologies and methods played a key role in enabling researchers to develop vaccines against COVID-19 in record time. But how can intellectual property rights keep up with the rapid pace of technological development?

This report seeks to take a holistic approach to copyright in the context of rapid technological change, as well as to suggest how a more agile and navigable balance between the interests of creators and the public can be achieved on the basis of two specific provisions of the InfoSoc Directive.

Starting point

The starting point for this text is that a Swedish government bill has been tabled to introduce the so-called three-step test as a provision in the Swedish Copyright Act.

When an author creates a work, they acquire exclusive rights to it. However, these exclusive rights are limited in certain cases where there is a strong public interest in allowing the public to use the work without authorisation or payment – such as for education or research. These are known as copyright exceptions and limitations. The EU's InfoSoc Directive contains an exhaustive catalogue of exceptions and limitations that may be introduced in national law. All of these limitations are subject to the three-step test, which makes it clear that works can only be used:

1. in certain specific cases (i.e. the regulated limitations);
2. the use must not conflict with the normal use of the work (i.e. compete with the author's ability to monetise the work);
3. and that the use must not unreasonably prejudice their legitimate interests.

The question, however, is how to apply this three-step test; a question that has been debated in the legal literature in Sweden and Europe for decades. There are two options that seem reasonable on paper. The first is to incorporate the three-step test into national law, thereby imposing broad limitations. Users know that they must comply with the test, have this to refer to, and case law becomes central to the detailed delimitation of the scope of the limitations. The second option is to actively apply the test when formulating national limitations, in such a narrow way that the three-step test becomes redundant in national law. This way, the user need not bother with the test and the courts need only apply it in unclear cases.

Copyright societies and collective management organisations have been fighting for decades for the incorporation of the three-step test into national law. However, they are also fighting for much narrower limitations. This means a double restriction of users' possibilities and rights. However, such a double restriction defeats the very purpose of the three-step test, which is to balance the interests of the author and the public.

Open norms

The problem with technology-specific legislation

In the Wikimedia judgment (NJA 2016 p. 212), he Supreme Court of Sweden discusses the exact meaning of the term "depict". The case in question concerns a database, Offentligkonst.se, which linked to digital reproductions of public art. The provision on freedom of panorama in Section 24 of Sweden's Copyright in Literary and Artistic Works Act allows depictions of public art in outdoor public places, but are digital representations depictions? The current wording of the panoramic provision was included when the Copyright Act was passed in 1960. It has been reviewed on a number of occasions, including in conjunction with the implementation of the InfoSoc Directive, without leading to any legislative changes. Swedish government report SOU 2011:32 called for further clarification of the concept of depicting, but this did not lead to any legislative action. Accordingly, the Swedish government's authors' rights committee's assessment that depicting refers to reproduction in the plane remains largely unchanged, but what is the exact meaning of reproduction in the plane when it comes to a digital environment? The ambiguity highlighted by SOU 2011:32 has persisted, as discussed in a legal inquiry from 2013.

The applicability of legislation over time is not necessarily a bad thing, quite the opposite. Future-proof, technology-neutral laws where case law can develop over time are a strength. However, this has not been the case with "depict", where the Supreme Court instead concludes that a restrictive assessment must be made along the lines of the three-step test. The first step is interpreted as "the limitation must be clearly delimited and specified", but it is not entirely clear whether this is an interpretation of the legislator's task or also an instruction to the courts. Clearer instructions to the courts are given regarding the second step, i.e. that the limitation should not compete with the normal exploitation of the work. In this case, the Supreme Court states that a dynamic interpretation should be made that "takes into account the author's right to exploit the work in the new ways made possible by technological developments," which in turn means that "the situation is different when the work is used in a digital environment." All the possibilities enabled by digital developments should, typically, accrue to the author.

If the law is always to be interpreted restrictively in the light of technological developments, this means either constant, further restrictions on the scope of limitations, as developments progress, or that the law needs to be updated for each major technological advance. Neither of these options is optimal.

The benefits of transparent legislation developed through case law

One alternative is to legislate instead for what has been called an open norm, whereby policy principles are established in statutory provisions and it is then left to case law to interpret these principles in the light of technological developments. A study by the Centre for Intellectual Property Policy & Management (CIPPM) at Bournemouth University defines an open norm as a "broad and non-exhaustive copyright exception, where its scope is flexibly determined and interpreted through a set of general criteria that is complemented at the level of the courts by a holistic assessment of legal, cultural, societal, and technological developments." The authors of the study analyse seven jurisdictions – the United States, Canada, Israel, Singapore, Sri Lanka, Japan and South Korea – where different variants of open norms have been implemented and examine the outcomes at the national level.

The study concludes that open norms have not only favoured the countries' creative sectors, but also their research and education sectors. The study also shows that open norms work in both common law countries (such as the UK and the US, where case law has a strong legal impact) and civil law countries (such as Japan and South Korea, where the law is based more on statutory rules), as well as in countries with some form of hybrid system, such as Canada and Israel. Based on the analysis of the seven countries, the authors of the study recommend that European countries also adopt open norms, so that copyright law can evolve with technological developments without all parties having to wait for lengthy legislative processes. The study also makes it clear that open norms are fully compatible with civil law-based legal systems.

The Supreme Court's preliminary ruling in the Wikimedia judgment already lays a foundation for a potential open norm. The ruling states that a "dynamic interpretation" should be made based on "the new ways opened up by technological developments", which in itself is an instruction to the courts to analyse the scope of the limitations based on the emerging technology. The major problem is that it also emphasises that what should be taken into account is "the author's right to exploit their works with the new technology." However, the balancing perspective, the public interest, is conspicuous by its absence in the Supreme Court's interpretation. The dynamic interpretation to be applied must presuppose that the author's right and the public interest – including the constitutional freedoms of expression and the press as well as academic freedom – are taken into account on the basis of technological developments.

InfoSoc Directive

One challenge for the introduction of comprehensive, open norms in the EU is the InfoSoc Directive, which was created in the early 2000s to harmonise European copyright in the emerging information society. The Berne Convention, which sets the framework for international copyright law, allows parties to the Convention in Article 9(2) to implement certain limitations in national law, provided they do not violate the three-step test (below). The three-step test has been interpreted in different ways in different jurisdictions and has been recognised by the WTO as being consistent with the US fair use norm, an open norm in which legislation establishes certain basic principles but the scope and application of those principles is determined by case law.

The InfoSoc Directive opted instead to establish an exhaustive list of permitted limitations. Article 5 sets out the twenty or so limitations that are permitted, but not obligatory, to be implemented in national law. This means that if a limitation is not included in the list or in subsequent EU directives, a member state may not maintain or introduce the limitation. It is true that Article 5(3)(o) allows for other limitations at national level, but only those that already existed at the time of the implementation of the Directive, which are of minor importance, and only in analogue form. Section 23.1, Paragraph 2 of Sweden's Copyright Act is an example of a provision in Swedish law that has been considered an implementation of 5(3)(o), and according to which one may reproduce public works of art for criticism, but only in analogue form (e.g. in a printed magazine). Thus, there is no possibility to introduce further limitations or exceptions in national law, unless it follows from a subsequent EU directive.

For most people, including the EU, it has become clear that the InfoSoc Directive's exhaustive list is not sufficient. For example, subsequent directives have introduced new mandatory limitations, such as the text and data mining (TDM) provisions of the DSM Directive, due to the InfoSoc Directive's lack of clarity in the light of technological developments. With technological developments, the limitation provisions of the InfoSoc Directive are no longer sufficient. It has also been argued that the InfoSoc Directive has failed to harmonise the internal market, has not taken sufficient account of users' needs for dynamic regulatory frameworks, has failed to anticipate the contribution of digitalisation to new uses, and has made it difficult for member states to respond to future technological developments.

The InfoSoc Directive's exhaustive list of specific, circumscribed limitations is a problem in itself for technology-neutral, future-proof copyright, but that does not mean that there are not good ways at a national level to exploit the Directive's potential. Below, two examples will be given of how national provisions under the InfoSoc Directive can be designed in a way that both fulfils the requirements of the Directive and allows for flexible overall assessments through case law. First, however, a few words on the three-step test.

The three-step test

The InfoSoc Directive's list of permitted exceptions and limitations ends with the previously mentioned three-step test (Article 5(5)). There are some important differences between the Berne Convention and the InfoSoc Directive's three-step test. The former focuses on the interests of the author, whereas the latter focuses on the rights holder, and it is increasingly common for copyright to be managed or administered by persons other than the author. In the Berne Convention, the three-step test is a general interpretative principle to balance the interests of the author and the public, whereas in the InfoSoc Directive it is more of a closed test to be applied to the already listed limitations, which in practice further limits their scope rather than being a general, balancing interpretative principle.

EU case law is not clear on the application of the three-step test, and the key concepts of the second and third steps – the normal exploitation of the work and the legitimate interests of the rights holder – have not been harmonised by the CJEU. This means that there is a lack of clarity as to how the test should be interpreted. There is some support for the Supreme Court's application in EU law, but also conflicting judgments that can be interpreted differently. Some judgments of the CJEU, including Funke Medien, suggest that the three-step test should be understood as an instruction to the legislator rather than to the courts, and that it is a matter of national implementation rather than individual application. The Supreme Court comes to the opposite conclusion, when it states in the Wikimedia judgment that the test should also be seen as an instruction to the courts to assess a particular application.

Since the Wikimedia judgment sets a precedent in its assessment of the three-step test, this means that as long as no other assessment is made by the Swedish legislator, the Supreme Court or the Court of Justice of the European Union, the three-step test will also be seen as an instruction to the courts. At the same time, the Supreme Court has only made limited use of the test in recent cases. Accordingly, there is no clear conclusion on its use in national law either.

According to the Swedish government report on copyright limitations (SOU 2024:4), the steps of the three-step test are cumulative. Reference is made, inter alia, to Stichting de Thuiskopie, case C-462/09, judgment of 16 June 2011, paragraph 31. owever, the interpretations of the CJEU are not unambiguous, and in recent cases – including Stichting Brein, paragraphs 63-71 – only some of the steps have been applied. According to Daniel Jongsma, Public Relations Consultants Association, case C-360/13, judgment of 5 June 2014 (paragraphs 54-62) is the only case where the first step has been applied, but on the other hand the third step is applied before the second. In other words, the three-step test is applied in reverse order, contrary to the report's formulation of the steps as cumulative. The conclusions that can be drawn from the PRCA and subsequent rulings by the CJEU – that the steps are not cumulative, that not all steps need to be applied and that they do not need to be taken in the given order – are in line with the declaration on a balanced three-step test from the Max Planck Institute, which Jan Rosén, among others, endorsed from a Swedish perspective. ater cases in Swedish law, such as the smartphone footage case, also clarify that while the three-step test can be applied by courts, it does not have to be applied by courts.

While there is little consensus on how to apply the three-step test, there is near consensus among academics and legislators that the three-step test does not need to be incorporated into national law. Some countries, such as France, Italy, the Czech Republic, Poland and most recently Belgium, have incorporated the three-step test into their national law, albeit in different forms, while other countries, such as Germany, the Netherlands, Denmark, Finland and Sweden, do not reproduce the three-step test. Regardless of whether the three-step test has been incorporated into national law, it applies: courts have invoked the three-step test in countries with the three-step test in national law as well as in countries without it, and of course also in the European Court of Justice. In many cases, the interpretations have been contradictory. The precise definition of its meaning is left to the European Court of Justice, which has also given conflicting interpretations.

In the event of national implementation, a number of starting points should be taken into account:

1. The first step of the three-step test, given the InfoSoc Directive's exhaustive list, is irrelevant from a national perspective and, as already mentioned, has been applied in very few cases in the CJEU.
2. When a limitation provision is limited to the specific case, in accordance with the first step of the three-step test, a judgment is made that use in these specific cases does not conflict with the normal use of the work, or in other words: a balance has already been struck during the legislative process between the economic rights of the author and the fundamental interests of the public. It should therefore be of less relevance to reproduce also the second step in a national perspective.
3. The three-step test should not limit the application of a limitation but should be used to determine whether a particular use falls under a limitation. In other words, it is not about limiting the normal case, but about balancing the rights of the author and the public in unclear cases. This is captured in the third step of the three-step test, where an overall balance is to be struck between the legitimate interests of the author and the public. The third step thus captures any ambiguities in the balance from the second step.

The most reasonable thing at national level, if an interpretation principle similar to the three-step test is to be introduced, is thus to reproduce the third step. This is also in line with the overall assessment that, according to both the CJEU and a long line of European legislators and academics, should be made. It is also in line with the PRCA, discussed above, which applied the third step first – the balancing of which also includes the second step. Such a variant is also found in other EU countries, including Ireland.

In order to ensure the balance to be struck, the provision, or at least the statute commentary, should, if introduced, also clarify that when applying the three-step test, the interests of the public, in particular the constitutional freedoms of expression, the press and academia, should be taken into account. In this way, the desired balance between the interests of the public and those of the author referred to in the Berne Convention will be achieved.

The advantages of a national implementation, properly designed, could be a greater scope for assessing the whole and the proportionality, where the limitations are interpreted on the basis of their purposes and objectives. This would also allow for a more flexible application, which is also in line with the fact that the CJEU has not always applied the test strictly and incrementally. The proportionality assessment and flexibility, with increased possibilities for courts to interpret the limitations in the light of the balance to be struck between the rights of the author and the public, also means that limitations can be imposed in a more directive-like manner, which also increases flexibility in relation to technological developments in society and the emerging case law of the CJEU. Properly designed, such a test can prevent legislation from becoming outdated, as with the Wikimedia judgment. Directive-based legislation, which can evolve in line with the emerging case law of the CJEU, also facilitates cross-border use and harmonises the internal market.

In her study, Raquel Xalabarder points to a trend whereby the CJEU's judgments are becoming increasingly flexible in an era of rapid technological change, and that the Court is also increasingly pointing to the need for harmonisation. A reproduction of the third step of the three-step test in national law could be an effective way to ensure that Swedish copyright law develops in line with the emerging case law of the CJEU.

In order to ensure that a national three-step test achieves the desired balance between different interests, the statute commentary should clarify that when applying the three-step test, fundamental freedoms, in particular freedom of expression and freedom of the press, should be taken into account.

Possible open norms in Swedish law

How could a national open norm work in practice? Below are examples based on two provisions of the InfoSoc Directive, Articles 5(3)(d) and 5(3)(h). The examples are not taken out of thin air but rather correspond to implementations of the InfoSoc Directive that have been made in other countries, including our Nordic neighbours in the first case and countries such as the Czech Republic, Ireland, Austria, Portugal and Germany in the second case.

The right to quote

The right of quotation is one of the few limitations already included in the Berne Convention. Influential academics such as Tanya Aplin (King's College) and Lionel Bentley (University of Cambridge) have emphasised that the right of quotation is a general limitation provision, and that the wording of the Berne Convention could even be seen as a global and mandatory open norm similar to "fair use": it is global because it is contained in the Berne Convention and the TRIPS Agreement; the wording of the Berne Convention makes it clear that it is mandatory; and the Berne Convention does not restrict it to any particular type of work, disposal or purpose, although names must be given when making available, and in accordance with fair use.

The right to quote (Section 22) is also one of the few general limitation provisions in the Swedish Copyright Act:

Section 22 Anyone may quote from published works in accordance with good practice and to the extent justified by the purpose.

Again, the legal provision itself refers generally to works, with the only requirement being that the works be published (which is also the case in the Berne Convention). In principle, therefore, it should be possible to quote all types of works, in line with international copyright law, as long as it is in accordance with good practice and to the extent justified by the purpose. The latter requirements mean that, for example, one cannot produce one's own works that consist only of quotations of other works, without the quotations requiring a context of some kind. However, as Swedish case law has developed, still images, photographs and works of art are not generally covered by the right to quote. This means that, as a rule, there is no right to quote photographs or works of art, regardless of whether they have been published and the use is justified by the purpose. This distinguishes Sweden from many other EU countries, including our Nordic neighbours. If you want to quote a work of visual art, you must instead use a different limitation provision in the Swedish Copyright Act, namely Section 23(1)(2) – which, however, only applies in analogue form:

Section 23 Published works of art may be reproduced in conjunction with the text in a critical presentation, but not in digital form.

When the InfoSoc Directive was implemented, the legislator considered that there was no limitation provision in the InfoSoc Directive that opened up for this type of use, and that the only possibility to preserve the limitation, which was already in the Swedish Copyright Act, was to use Article 5(3)(o):

5(3)(o) Use in certain other cases of minor importance where exemptions or limitations already exist under national law, provided that they only concern analogue uses.

However, this is a strange interpretation, as the provision seems to be fully in line with Article 5(3)(d) of the InfoSoc Directive:

5(3)(d) Quotations for purposes such as criticism or review, provided that they relate to a work or other subject-matter which has already been lawfully made available to the public, that, unless this turns out to be impossible, the source, including the author's name, is indicated, and that their use is in accordance with fair practice, and to the extent required by the specific purpose.

There is nothing in Article 5(3)(d) that limits the provision to excluding visual works and photographs. Rather, the development of the law in the Court of Justice of the European Union has moved in a direction where it is clearly stated that it is possible to quote photographs and works of visual art. The so-called Painer judgment (C-145/10) assumes that photographic works are covered by the limitation (and neither the European Commission nor the countries that expressed their opinion in the case expressed a different view). Another similar interpretation is found in the Pelham judgment, according to which quotations can be made regardless of the type of work, as long as there is a "dialogue" with the quoted work:

71. As regards the usual meaning of the word 'quotation' in everyday language, it should be noted that the essential characteristics of a quotation are the use, by a user other than the copyright holder, of a work or, more generally, of an extract from a work for the purposes of illustrating an assertion, of defending an opinion or of allowing an intellectual comparison between that work and the assertions of that user, since the user of a protected work wishing to rely on the quotation exception must therefore have the intention of entering into 'dialogue' with that work, as the Advocate General stated in point 64 of his Opinion.

A growing number of authors on the subject also draw the interpretation that Pelham and Spiegel Online contribute to introducing an EU-autonomous definition of quotations, i.e. one that is given the same interpretation throughout the EU and independent of definitions at national level. This definition also includes photographs and still images.

No other Nordic country has restricted the reproduction of works of art in critical representations to analogue uses only, and Painer, Pelham and Spiegel Online make it clear that the right to quote also extends to images and photographs.

The European view that images are included in the right of quotation, including for criticism, is made clear in the DSM Directive. Section 52 p of the Swedish Copyright Act currently provides a right for users of information society services to quote works, without limitation to works of art, on the platforms covered by the provision (defined in Section 52 i). This means that there is currently a right to quote works of art on digital platforms covered by Section 52 i, as well as in analogue form under Section 23, Paragraphs 1-2. However, this still excludes a number of digital platforms, including the Wikimedia platforms but also many mass media. This is despite the fact that, according to the case law of the Court of Justice of the European Union, mass media reporting to the public is recognised as having an even greater value in terms of freedom of information and expression than digital reproductions by individual users.

The Swedish system seems incompatible with European case law. The most reasonable solution would be to strike the requirement that reproductions may only be made in analogue form in Section 23(1)(2) and to leave the detailed meaning of the general limitation on quotations (as implemented by Sections 22 and 23) to the courts for balancing the interests of the author and the public. Such a solution lays the foundation for an open norm, where case law defines the detailed trade-offs to be made in the balance between the interests of the author and the public.

Panoramic freedom and a dynamic interpretation of technological developments

The case in Swedish law where the Supreme Court has applied the three-step test and set a precedent concerns the so-called panorama exception, Section 24, Paragraph 1 of the current Swedish Copyright Act, which gives everyone the right, without the author's permission, to depict public works of art that are permanently placed in a public place. The Supreme Court judgment thus states that the freedom of panorama only covers reproduction in the plane, which has been interpreted by many as two-dimensional reproduction and not digital. However, the precise meaning is unclear: after the judgment, is it illegal to reproduce images digitally at all, or only in the form that took place via the Offentligkonst.se project? In other words, is it prohibited to share images of public art digitally, even for private individuals (as the Swedish Patent and Market Court suggests in its subsequent judgment), or is it prohibited to compile a structured database that links to digital reproductions of public artworks? The lack of clarity is a major problem for digital actors. Volunteers on Wikipedia have stopped documenting public art, which is a major shortcoming for education, public awareness and democracy, and it is highly unclear whether public art can be shared on social media.

The Swedish exception is in several ways more restricted than the InfoSoc Directive. For example, it covers only works of art and not all types of works, and only works placed outdoors. The question is to what extent these additional restrictions fall within the national legislator's discretion, that is, within the framework of how far the national legislator may diverge from the text of the directive. However, this is a discussion that the Supreme Court does not take up, instead discussing the precise meaning of the concept of depict. The term is not defined in Swedish copyright law, despite the fact that Swedish government report SOU 2011:32 called for a clearer definition. Based on its interpretation of the three-step test, the Supreme Court concludes that "a dynamic interpretation that takes into account the author's right to exploit the work in the new ways made possible by technological developments " should be made. The Supreme Court argues that there is indeed a strong public interest in this case (paragraph 21), but that it is irrelevant because, on the basis of the dynamic interpretation, it may entail a financial loss for the author. The "right to exploit works by means of new technologies" therefore remains with the author.

The Wikimedia judgment does not really balance the interests of the author and the public. Despite the high public interest, a restrictive interpretation is to be made, in what appears to be a reference to the three-step test. However, this interpretation is not clearly supported by EU case law, and in a case such as England And Wales Cricket Board v Tixdaq (2016), the Court is clear that the interests of the author must be balanced against those of the public:

Rather, the exception can apply unless those interests are unreasonably prejudiced. This requires consideration of proportionality, and a balance to be struck between the copyright owners' legitimate interests and the countervailing interests served by the exception.

The academics who signed the Declaration on a Balanced Interpretation of the Three-Step Test also emphasise that this balance between the interests of the author and the public is central to the application of the three-step test, but this is not taken into account in the Wikimedia judgment.

The three-step test opens up for development via case law where new assessments are made based on technological developments. However, the Supreme Court's preliminary ruling means that all technological developments typically accrue to the author. This in turn means that the public interest is eroded for every technological innovation, in itself a severe restriction on freedom of expression and other constitutional rights, restrictions that also discourage innovation and development.

A three-step test implemented in such a way as to balance the interests of the author and the public on the basis of the new, potentially ambiguous cases opened up by technological developments provides an opportunity for case law to fill in where the legislator cannot. However, this requires provisions that are broadly formulated and closely implement EU rules. Otherwise, the three-step test will become a double restriction rather than a balancing rule of interpretation, and there is no support for such a use of the test in either European or international law.

A basis for an open norm can be found in Article 5(3)(h) of the InfoSoc Directive, which states that works, without defining exactly which ones, which are permanently placed in a public place can be used freely. This provision of the InfoSoc Directive opens the way for development through case law, where the precise limits of the scope of the limitation will be determined by the courts, inter alia in the light of technological developments. However, this requires both a provision in Swedish law that is close to the Directive and that the three-step test is applied in a way that helps to strike a balance between the interests of the author and the public.

Conclusion

Technological developments are moving very fast, and digital technologies are opening up new ways of using works. However, these developments pose challenges to the ability of authors to monetise their works, while at the same time creating grey areas where it becomes unclear which limitations actually apply.

For a number of reasons, there are good reasons to implement the limitations in the EU directives as closely as possible. Firstly, it means further harmonisation of EU copyright, which promotes opportunities to both protect and share works within the EU, as highlighted by the Confederation of Swedish Enterprise in its report "Globalisation in the knowledge economy". On the other hand, it means that case law can fill in where new opportunities arise, without the need to constantly rewrite the law – which is a slow, cumbersome and expensive process.

However, the Supreme Court's ruling in the Wikimedia case has helped to create new grey areas, with an unclear and controversial application of the three-step test that makes it unclear how to use protected works with new digital technologies.

Implementing limitation provisions close to the Directive, complemented by a three-step test essentially based on the third step of the three-step test, opens up the possibility of letting courts decide how to assess new cases based on technological developments. Such a three-step test should clarify that when applying the three-step test, fundamental freedoms, in particular freedom of expression, freedom of the press and academic freedom, are taken into account. This ensures that the rights of authors are protected, in that uses that unreasonably interfere with their legitimate interests are not permitted. But it also ensures that the interests of the public, such as for research, education and training, are protected.

The examples above illustrate how open norms could be implemented in a way that protects the interests of both the author and the public while allowing the courts to strike the right balance in the light of technological developments. Such a regime would benefit creators, users, innovation, research and – ultimately – both the economy and democracy.

Open science and open access

At the global level, we are seeing extremely rapid progress towards open science and open access. UNESCO has adopted a global recommendation on open science, and during the Swedish EU Presidency, issues concerning open science were given a strong European flavour through the strong proposals included in the Council Recommendation. Country after country in the EU is legislating on so-called secondary publication rights (SPR), which allow researchers to publish their research results in parallel in open research databases regardless of the policies of commercial publishers. At national level, the Swedish government emphasises the importance of open science in its research bills, and major national research funders require the open publication of any research they fund. UNESCO's recommendation on open science has also prompted Sweden's new national guidelines for open science, developed under the leadership of the National Library of Sweden.

These developments are essential to promote the dissemination of open knowledge. Open science, with better opportunities for researchers to make their research freely available, is essential for the dissemination of knowledge in society, and fundamental to projects such as Wikipedia and other popular education projects, which collate and make knowledge available to the public. Open science also helps the Wikimedia platforms to assist in the third task of bringing research and knowledge from colleges and universities to the public. It is also, in co-operation with the public that shares and makes it available, an important counterforce to the spread of disinformation and inadvertent misinformation.

However, as strong as the support for open science is, many copyright barriers to research still remain. These barriers apply to both conducting and disseminating research in an efficient, modern and innovative way. They also prevent researchers and the public from taking advantage of the opportunities that digital developments open up for the dissemination of knowledge and the utilisation of research at societal level.

It is these barriers that this report will highlight. It attempts to take a holistic view of the academic process, highlighting the barriers to both research production and dissemination, based on the idea that, in general, a copyright environment that favours research is necessary for research to benefit society in a free and accessible format. If legislation discourages the academic process in any part, it has a knock-on effect on the whole process, and therefore permissive legislation in all parts is necessary if the goal is to make science open, accessible and free. This includes future-proof, technology-independent legislation that favours, not hinders, digital opportunities.

International outlook

In the UNESCO Recommendation on Open Science, member states emphasise that while open science is based on intellectual property rights, it also requires open licences and flexibility in copyright to allow access to knowledge for both researchers and the public, as well as to create opportunities for innovation and citizen science. As several European studies and reports have shown, inflexible copyright frameworks at European and national level effectively discourage innovation and development, thereby contributing to a loss of European competitiveness at global level. This is also true for non-profit actors such as the Wikimedia platforms, whose full potential cannot be realised when copyright law imposes barriers to innovation.

In the light of the UNESCO recommendation, the European Council, led by Sweden, adopted recommendations "welcoming" the implementation of secondary publication rights in national law and mandating the European Commission to remove barriers to research as part of the second item of the European Research Area policy agenda. However, the copyright challenges for open science and research are many. At the European level, one of the major challenges is the high level of copyright fragmentation. The so-called InfoSoc Directive was supposed to help harmonise copyright in the information society, but there are still major differences between EU member states in terms of copyright conditions for research. The provisions of the InfoSoc Directive have been implemented in very different ways, often involving cherry-picking through partial implementations or rather discretionary interpretations. As a result, copyright legislation in the EU often resembles a patchwork quilt, with major differences in how the provisions are formulated between member states. Although in many cases this has been necessary, not least because of the development of the digital society, it has made it almost impossible to follow how the InfoSoc Directive has been implemented at national level, and to know which national provisions contradict the EU Directive. This in turn means that it is very difficult for Swedish researchers, for example, to know how they can collaborate with researchers in other EU countries, or what the legal conditions are for accessing material from other EU countries.

At European level, there are proposals to make knowledge a fifth freedom of movement (after goods, services, people and capital). The basic idea is that knowledge must be able to flow freely in the knowledge society, and that national borders for knowledge are an obstacle to the realisation of the internal market. This is the argument of former Italian Prime Minister Enrico Letta, for example, in the so-called Letta Report of April 2024. However, the road to such a fifth freedom of movement may be long, and until then there is much Sweden can do at national level to strengthen the opportunities for open science. Below are proposals aimed at making it easier for researchers to communicate research, for the public to access research, and for organisations to build a strong national research infrastructure.

The Wikimedia movement is a global movement. The barriers created by national political boundaries effectively counteract the ability of Wikimedia volunteers to disseminate knowledge freely. Thus, the lack of harmonisation in the EU and globally has negative consequences not only for researchers, but also for the exploitation of research results that is necessary for research to benefit the public. A good research environment in Sweden, where barriers to the dissemination of knowledge within the EU are removed, is a prerequisite for reliable knowledge on Wikipedia – one of Sweden's most widely used websites that the public uses to access reliable and qualitative knowledge.

Use of works in scientific research

Several EU directives, and in particular the 2019 DSM Directive, emphasise the great value of research, and a copyright law that promotes and does not hinder research purposes. Several recitals of the DSM Directive, including the fifth, discuss the fact that the InfoSoc Directive, which was created in the early 2000s, at the dawn of rapid digital development, does not in itself allow for several important uses that digital technology has since made possible. This, in turn, puts the EU at a competitive disadvantage at the global level. Accordingly, there is a realisation at EU level that European legislation puts European researchers at a disadvantage compared to their colleagues in other parts of the world.

However, several EU countries have implemented the broad provisions of the InfoSoc Directive even before the new provisions of the DSM Directive, to provide a sound national copyright basis for research. For example, the UK, while still in the EU, was an early implementer of TDM provisions, based on the InfoSoc Directive's limitations, and similar solutions have been in place in several countries even before the implementation of the DSM Directive. TDM and other digital research methods were used before the DSM Directive, but in the absence of national and European policies, they operated in a legal grey area, which the DSM Directive helped to clarify.

However, the two TDM provisions transposed into Swedish law by the DSM Directive only allow for the making of copies, i.e. copying of works. They do not provide for a right of communication to the public, which is necessary not only to allow researchers remote access to research material, but also to share research results and data for verification and validation as well as for dissemination of the results to the public. The lack of provisions for transmission to the public creates major obstacles to transparent research and prevents researchers from working in accordance with the requirements of open access to scientific research. It also creates major problems for collaborative and transnational research, as well as for the public utilisation of research results. This applies to all research involving copyright-protected works.

It can be difficult, if not impossible, to legally anticipate the future needs and obstacles of research. Legislative processes take time, so it is important to have a technology-independent, future-proof law that favours rather than discourages innovation. To this end, the best solution is to introduce broad limitation provisions based on the InfoSoc Directive into national law, allowing courts to interpret, based on the three-step test and statutory principles, whether uses falling outside the limitation provisions are reasonably balanced between different interests. One such provision is Article 5(3)(a):

(a) use for the sole purpose of illustration for teaching or scientific research, as long as the source, including the author's name, is indicated, unless this turns out to be impossible and to the extent justified by the non-commercial purpose to be achieved

National provisions based on 5(3)(a) were implemented in several countries in the context of the implementation of the DSM Directive, and similar solutions already existed in several EU countries. In the absence of a limitation provision in the InfoSoc Directive or subsequent EU legislation, the provision may not be introduced at national level. A general article such as 5(3)(a) has therefore been central to the implementation of the necessary copyright limitations for research purposes in various EU countries.

In its review of copyright barriers to research in the EU, DG RTD notes that Swedish research limitations are more restrictive than in the rest of the EU. This has a tangible impact on the everyday lives of researchers and counteracts Swedish ambitions to be a leading country for research and innovation. It also makes it more difficult for researchers, companies and the general public to contribute to the utilisation of research and the dissemination of research results. To remove obstacles to Swedish research, a broad, general provision in line with Article 5(3)(a) of the InfoSoc Directive could be introduced, for the making of copies and transfer to the public for research purposes. The reason both uses are necessary is that they are both inseparable and essential parts of the research process.

Article 5(3)(a) of the InfoSoc Directive has been interpreted as compatible with the first step of the three-step test by the EU and fulfils the second step by clarifying that the use may only be made for the non-commercial purpose to be achieved. Thus, while having a restricted impact on the economic exploitation of authors (the normal use of the work), it would drastically increase the possibilities for researchers (who are also authors) to explore protected material, to the benefit of both research and the public. A national implementation of 5(3)(a) close to the Directive, together with a general principle of interpretation based on the third step of the three-step test in national law, would promote research at national level and be compatible with the three steps of the three-step test. The precise boundaries would be left to the courts, in line with the requested flexibility and proportionality assessment, and in order to strike a balance between the rights of the author and the public.

It is essential that researchers are able to conduct research on all types of protected works. To foster collaborative and transparent research, researchers must also be able to copy, reproduce and share protected works, in line with the FAIR principles.

Reproduction of works for scientific purposes

A closely related issue to the use of works for scientific purposes, in the sense of creating opportunities for researchers to collaborate and access materials, is the reproduction of protected works in research results, such as in academic articles. For research to be accessible, understandable and illustrative, there must be ample opportunities for researchers to reproduce protected works in articles, monographs and other academic contexts.

Section 23(1)(1) of the Swedish Copyright Act contains a limitation that allows works to be reproduced for scientific purposes, but only in non-commercial contexts. This is an important provision that is central to being able to reproduce works in relation to scientific contexts and is a partial implementation of Article 5(3)(a) of the InfoSoc Directive. This means, for example, that in a scientific article on an artist's subject, a researcher can reproduce works of art by the artist to illustrate these subjects, but also, for example, for the reproduction of graphs and diagrams.

The non-commercial restriction comes from the InfoSoc Directive and is thus unavoidable. At the same time, it creates complicated grey areas. Research funders in Sweden and the rest of Europe require open licences, usually CC BY, for research results that they fund. At the same time, the CC BY licence opens up for commercial reuse. Where is the limit for when CC BY on research results means that the use falls outside the scope of the limitation, if the work is reproduced on the basis of Section 23(1)(1) of the Swedish Copyright Act? To what extent can the limitation be used to reproduce public works in research published in open-access scientific journals, which generally require publication under CC BY licences (and which allow for commercial reuse), especially if the journal is run by a commercial publisher? To what extent can works be used for research purposes if a company is involved in the research project? In particular, if the company also has an interest in reusing the results of the research project? A limitation restricting such use would run counter to the Swedish government's research policy bill as well as the national guidelines for open science, and few Swedish research funders allow publication with anything other than CC BY. It also hinders the dissemination of free knowledge in society, and all those non-profit, educational or popular education actors – such as Wikipedia and the Wikimedia movement – that use open licences to clarify conditions for the reuse of knowledge, thus promoting knowledge dissemination.

The wording of Article 5(3)(a) and Section 23 of the Swedish Copyright Act means that European researchers involved in public-private partnerships do not have the same opportunities to benefit from the results of their research as researchers in other parts of the world. This in turn is a major obstacle to the utilisation of research in both Sweden and Europe. In a Swedish context, this counteracts the efforts made by the Swedish government to enable Swedish government agencies such as Vinnova, the Patent and Registration Office and the National Agency for Public Procurement to promote the utilisation and commercialisation of research, thereby strengthening Swedish innovation and competitiveness. KR21 has called this the EU's "innovation paradox", and it is an issue that is also raised in the study from DG RTD cited above. The commercial restrictions in the European limitations on the use of research are counteracting European innovation and competitiveness globally.

There is nothing Sweden can do in national legislation to change this, but Sweden should pursue the issue at EU level of relaxing the hard boundary that has been created between commercial and non-commercial research. The EU's research strategies emphasise the importance of collaboration between universities and commercial actors, while copyright law undermines precisely this type of collaboration. It is also a major obstacle to the dissemination of free knowledge, including through digital projects such as Wikipedia.

Remote access

One key issue when it comes to creating favourable copyright conditions for research is remote access, whereby researchers can access research materials remotely, without physically visiting the institution holding the material in its collections. This issue is addressed in Swedish government report SOU 2024:4, which states that the issue is a matter for those concerned rather than the legislator, with the conclusion that no changes are needed to strengthen these possibilities. This is after it has been noted that an institution may only be allowed to make copies for the use of its own institution, and that in several cases researchers have had to digitise material themselves and then destroy it, even though it has already been digitised at other institutions. While there are different views on the possibilities for remote access provided by the law, it is clear that legal grey areas remain.

One such ambiguity is the meaning of 'lawful access', which is a requirement for copying. The European Commission's review of copyright barriers to research in the EU states that the lack of clarity about what the term means constitutes a major obstacle to researchers gaining access to research material. It is further stated that it is a matter of urgency to correct this ambiguity and that it should be investigated how Article 5(3)(a) of the InfoSoc Directive can be used to give researchers access to material. There are no EU legal restrictions on this, and through the DSM Directive, secure electronic environments have become an EU legal concept. However, the study also points out how a general limitation under Article 5(3)(a) provides the necessary legal basis for researchers to access material for their research, without having to digitise material that has already been digitised, with all the costs that this entails. It also makes the research environment fairer across Sweden, as researchers at all higher education institutions, regardless of where they are located geographically, are given equivalent opportunities to research relevant material by being able to access digital material remotely.

At a societal level, the lack of clarity regarding remote access leads to major inefficiencies in both economic and environmental terms. In a large, dispersed country like Sweden, the most efficient use of resources for building research infrastructure would be for one large national institution (such as the National Library of Sweden) to have overarching responsibility for building the infrastructure itself, but for this infrastructure to be available for use by researchers throughout Sweden. Rather than having to physically visit the legal deposit libraries, researchers often choose to research other materials instead, materials to which they have easier access, which means that copyright rather than research needs influence what research is pursued. In many other contexts, researchers choose instead to make their own copies of works for their research, which means that the same work can be digitised several times. Remote access would save emissions from many thousands of journeys, which would be an environmental benefit, but above all, it would make research more efficient and allow researchers to devote their time to what is valuable: research.

This is a major cost to society in itself, and we can add to this the fact that this material must then be stored on secure servers within each institution that digitises the material, which of course also comprise major costs. Much material within the institutions is unique to each institution, but material that has already been copied and stored at one institution should be available to researchers at other institutions to ensure that limited resources are used wisely. However, a prerequisite for this to be possible is legislation that enables institutions and organisations to share material outside institutions, including by implementing a broad limitation in line with Article 5(3)(a).

Works not available on the market

Another issue related to remote access comprises the two new provisions for works that are not commercially available. One is a limitation that allows works to be digitised and made available to the public if they are not commercially available, but only in cases where there is no collective management organisation mandated to represent such works. If there is such a collective management organisation, the works can instead be made available on the basis of a new contractual licence, which is also included in Swedish copyright law following the implementation of the DSM Directive.

Much of the literature requested as interlibrary loans from research libraries and similar institutions is not commercially available, especially since commercially available works are usually also available in many libraries. There is a strong public interest in making works that are difficult to access more easily available. Today, however, this can almost exclusively be done through interlibrary loans, where books and works are physically sent between libraries. Instead of these works being sent as physical interlibrary loans between libraries, the works could be digitised and made available via non-commercial platforms, according to the new limitation or contractual licence for works that are not commercially available.

However, the collective management organisations have not yet developed licences that cultural heritage institutions can use, and at the same time there are collective management organisations representing almost all types of works in Sweden. This means that there is a direct obstacle to the possibility of making works available that are not commercially available. How effective this barrier is at present becomes apparent if one looks at the digitised works not commercially available that have been reported to the EUIPO web portal: about one hundred Swedish-language works are registered in the portal, but none of the works have been registered by a Swedish institution. No works at all in Sweden appear to have been made available under the limitation or the contractual licence.

The DSM Directive requires that the government initiate consultations to bring together interested parties, in turn to promote the development of a functioning licensing market for non-commercial works. However, the government has not done this, and thus does not fulfil the requirements of the DSM Directive.

If the provisions were to be applied, it would help to make works that are difficult to access available to researchers, but also, of course, to members of society who want to engage in private study or education in any field. Writers on Wikipedia, for example, would have much easier access to source material, which would benefit society as a whole. Greater access to source material enriches and contextualises the knowledge on Wikipedia. Moreover, works that are not commercially available could thus indirectly become an important source for the production of free knowledge. At the same time, the economic loss for authors and creators would be negligible, as the provisions can only be applied to works that are not available through commercial channels.

Technical protection measures and contractual overrides

Two major obstacles to research in Sweden are so-called technological protection measures (TPMs) and the possibility to override limitations by contract. Technological protection measures involve digital technologies to prevent access to and dissemination of protected works, technologies that are permitted to be used except in certain cases. Contractual overrides involve negotiating away certain legal rights to use the material through contracts, such as in subscription services for access to material. When the DSM Directive was implemented, Swedish law needed to be adapted so that, for example, researchers were given the right to circumvent technological protection measures for TDM purposes (in the context of research), and that the rights opened up by the TDM limitation for research cannot be overridden by contract. However, this only applies to the TDM limitation for research and not to other research.

Technological protection measures are governed by the provisions of Sections 52 d and f of the Swedish Copyright Act. Section 52 d stipulates that the protection measures may not be circumvented without the consent of the author or rights holder. However, to enforce this right under Section 52 f, the user, such as the researcher, must go to court, which can be a costly and time-consuming process. It also means that in many cases researchers are unlikely to use certain material that they have a legal right to access because the process of asserting their right is too cumbersome. The current provision makes it difficult for researchers to defend their rights.

The right to circumvent technological protection measures also applies only to the one TDM limitation (for research). It does not apply to the general TDM limitation. Nor does it apply to the limitation on making copies for research purposes by research organisations and cultural heritage institutions, but only for preservation. So organisations can make copies for preservation purposes, even if this means circumventing technological protection measures and regardless of what the contracts say, but not for research other than TDM. The TDM provision for research is also the only research provision that cannot be contractually removed. This means that contracts can restrict the right of research organisations and cultural heritage institutions to make copies for research purposes.

It should be made easier for users, such as researchers, to assert their rights to access material. A review should also be carried out to better balance the author's interest in the non-dissemination of their work with the public's interest in the free dissemination of knowledge. It seems unreasonable that cultural heritage institutions can claim their right to copy works for preservation but not for research purposes. The law should also make it clear – in line with the new limitation in Section 49 a of the Swedish Copyright Act, which states that new copyrights cannot be claimed on reproductions of works where the copyright has expired – that technological protection measures are not allowed on copyright-free material.

Secondary publishing rights

It is important to improve the conditions for researchers to conduct their research, to be able to access material remotely and to share it with research colleagues. Researchers must also have good opportunities to use works in their research as well as to disseminate research that reproduces protected works to the public. However, it is also crucial that academic authors have good opportunities to publish their research openly.

Research funders in Sweden as well as the Swedish government require open publication of publicly funded research. Open access to research is central to counteracting the spread of disinformation and promoting source criticism and media and information literacy. Easily accessible research is an effective way to increase the level of knowledge in a society, and for the volunteers of the Wikimedia movement, it is a prerequisite for being able to contribute with qualitative knowledge via, for example, Wikipedia.

However, researchers are constantly facing challenges when trying to publish openly. For example, they get caught up in tricky contracts with commercial scientific publishers. While they want to publish openly, publishing in prestigious journals is important for their credentials, and while some of these journals allow open access, many have contracts that are difficult to understand and that, to varying degrees, sign over the author's rights to the journal. The lack of legal and juridical clarity puts the author in a difficult position, in many cases having to choose between the funder or the scientific journal due to conflicting requirements.

The ERA policy agenda emphasises the importance of raising researchers' awareness of their rights and how to protect them when dealing with publication agreements. Several studies and reviews have highlighted the intricacies of navigating academic journals' publication agreements, and how difficult it is even for lawyers to understand the finer points of the agreements. At the same time, it is clear that simply raising awareness is not enough, and legislative action is needed to realise national goals of open access to science and to make it easier for researchers to publish their research openly.

The most important such legislative change, highlighted in the Council Recommendation of the Swedish Presidency of the EU, is secondary publishing rights (SPRs). SPRs ensure that authors of academic material have the right to publish their material openly, in a non-commercial scientific database, even if they also publish the material closed in a commercial journal. Accordingly, this allows for parallel publication, where a particular version of the research results can be published in parallel with the version printed in the journal.

SPRs have been introduced in almost a dozen EU countries. There are different ways of constructing the provisions. They can be implemented as a copyright limitation, as Christina Angelopoulos, among others, points out in an EU study. However, most countries have chosen to implement SPRs as a limitation on freedom of contract. A further possibility could be to legislate on SPRs as part of the right of first publication in Section 8 of the Swedish Copyright Act.

Either way, an SPR, especially if complemented with an obligation to publish in open file repositories, would have great value in society and for the utilisation of research. It would increase the amount of reliable, qualitative and free knowledge in society, and make it easier to access and share information even for those with limited resources – both in Sweden and globally. For the Wikimedia movement, this would mean extremely important access to source material, where volunteers on Wikipedia, for example, could contribute to the third task of colleges and universities, to bring research to the community, and to update Wikipedia more rapidly with high-quality sources. This will of course be particularly important in controversial topics, crises and disasters, or in niche areas where there are otherwise few other quality sources available.

Conclusion

At Swedish and European level, research, open science and open access to science are highly prioritised areas in political strategies, bills and research budgets. Increasingly more research funders are also demanding open publishing, usually with a CC BY licence. However, the review presented above shows how Swedish and European legislation in many cases actively counteracts the openness that is the policy's ambition. This applies to the opportunities for researchers to use protected works in their research and to access the works remotely, as well as to the possibility of circumventing technical protection measures in order to research material. When material is difficult to access, there is a risk that research results will be inferior and that researchers will choose research based on what material they can access rather than what material would be of societal and academic relevance. This is a major problem that should be prioritised at the political level, such as through a statutory right for researchers to access material for their research.

In addition, there are major challenges for researchers to openly share the material they themselves have authored. Complicated legal publishing agreements and a weak bargaining position with meritorious, prestigious journals mean that researchers, unwittingly or because of the difficult situation, lock up research despite the demands of policy and funders. In order to strengthen the position of researchers and to provide statutory rights to publish research openly, secondary publishing rights (SPRs) should be introduced into Swedish law, as almost a dozen other EU countries have already done. In some of these countries, a parallel publication obligation has also been introduced – in line with the fact that publicly funded research shall also be publicly accessible.

Open access to science is fundamental to promoting the dissemination of knowledge in society. It is an effective tool against disinformation and an important part of fulfilling the third mission of higher education institutions. There are rapid developments at policy level to promote the open dissemination of science, supported at global level (e.g. by UNESCO), at European level (e.g. as reflected in the Council Recommendation on Open Science from 2023) and at Swedish level, through the national guidelines on open science and the government's research policy bill. At the same time, developments have not progressed as far at the legal level. This report has identified a number of necessary legislative changes to promote open science at all levels.

When researchers have good opportunities to access material, research material, share research material with their colleagues, share material in their research results, and then disseminate those research results openly, society also has good opportunities to access qualitative research, and with the help of SPRs and other legislative measures for open access to contribute to the dissemination of free knowledge in society. The latter via free knowledge projects such as Wikipedia, for example.

TDM & AI

The academic world is currently undergoing rapid digital transformation. New technologies are opening up new ways of conducting research while digital tools are making it easier for researchers to work remotely and across borders.

The DSM Directive implemented two new provisions for text and data mining (TDM), one of the most high-profile of the new digital research methods. In short, TDM involves the use of computer-based methods to conduct research on very large numbers of works, including analysing trends, patterns and correlations. The results of TDM can have a very wide range of applications. While the debate focuses on generative AI, TDM and AI development have a major impact on science, research and competitiveness, and in areas as diverse as healthcare, agriculture and the environment. For the Wikimedia movement, the provisions have a major impact. The platforms, in particular Wikipedia, are an important source of material for TDM, including large language models (LLMs). In addition, material from the Wikimedia platforms has a significant impact when reused in other services that utilise AI for selection and presentation, such as search engines.

At the same time, it has quickly become clear that copyright has not kept pace with technological developments. Many grey areas have emerged, making it difficult, if not impossible, for researchers to know what actually applies and how they can use these new technologies. The AI Regulation, which was adopted in spring 2024, does not address any of these issues, so for the foreseeable future it will fall to the national legislator to sort out the grey areas.

The new provisions only concern the making of copies, the right to copy works for TDM; they say nothing about the possibility of making the copies available for any purpose. In addition, the research provision focuses on institutions and organisations rather than researchers, making the relationship between researchers and institutions unclear when using works. This may mean, for example, that researchers must be physically located at the institution that holds the material and thus have no possibility of accessing it remotely, that the researcher may not share the material with research colleagues or for verification purposes, either in Sweden or abroad. This will also make citizen science and public involvement in the research process difficult, and it is unclear from a copyright perspective to what extent the copies may be stored – despite the requirements of the Swedish Archives Act.

This report attempts to determine, step by step, the relevant actors and how they are hindered by the new rules.

The TDM process

Many AI models, such as LLMs, analyse and generate text based on patterns they have identified in, and learned from, very large datasets. This AI training typically starts with the collection of these datasets, which can consist of web pages, scientific or journalistic articles, books, images and other written or visual information available on the internet. For example, Wikipedia articles are a common source of TDM input.

This data is then cleaned up, sorted and annotated, with the cleanup process being done manually or automatically. Once the data has been cleaned up, it is fed to the model to be developed. The model uses algorithms to identify patterns, trends, relationships, correlations and structures in the data. Based on these patterns, the model adjusts and improves its own parameters throughout the process, thereby enabling the model to start predicting words or answering questions based on the data it has been fed. The more data the model is fed, the more accurate the answers it can provide.

Data collection can be done by a web crawler that collects material from different websites, text and information that is stored and used to train the models. More niche models, such as for research purposes or specific application areas, can be based on specific corpora, developed for the specific purpose – for example digitised books from a library. The reason one usually also wants to train these models on copyrighted material, such as books or articles, is that they contain necessary descriptions of, for example, emotions, language nuances and relationships that the model itself cannot perceive. Advanced data is needed for advanced AI.

TDM and copyright

AI training poses two copyright risks: it can lead to unauthorised communication to the public and unauthorised copying (i.e. illegal copies), two rights typically reserved for the author. The first risk, such as an AI model delivering copyrighted material as output, is relatively easy for models to address. One can limit the responses that AI models provide with this in mind, much in the same way that models stop inappropriate language or incitement to violence. Avoiding copying is more difficult, as the material is copied several times during the training process. For example, data (such as text and images) needs to be copied when it is collected, when it is cleaned up and annotated, and when it is calibrated. Prolonged limitations on the possibilities for TDM copying risk increasing bias and thereby reducing the reliability of the results. The focus should therefore be on output rather than input.

The InfoSoc Directive has a mandatory limitation provision (Article 5(1)) for the making of temporary copies, Section 11a of the Swedish Copyright Act. The limitation is necessary to make it possible, for example, to watch or listen to a work online, as it authorises the temporary copies that are created in various internet servers to make this possible. However, the limitation is not restricted to this, but allows all temporary copies "if the preparation of the copies constitutes an integral and essential part of a technological process and if the copies are transient or have only a secondary importance in that process." The Infopac case clarifies that transient copies arising from the collection and processing of data are permitted. The judgment suggests that certain forms of data collection and processing that occur in the development of AI models are permitted. However, this has not been tested in Swedish law.

However, since 1 January 2023, there are two new provisions for TDM in the Swedish Copyright Act, Sections 15 a and b, which provide a legal basis for making copies for TDM purposes. While Section 11 a potentially allows some elements of AI development, particularly in relation to the fact that facts and ideas are not subject to copyright, the new provisions add clarity, although risks and limitations remain.

The first provision contains a general limitation, giving everyone the right to copy works for TDM purposes. However, this is restricted by the fact that authors and rights holders can make reservations against this use. Accordingly, there is the possibility of opting out, so that users cannot use the authors' works under the provision. However, on user-generated platforms, such as Wikipedia, or other platforms using open licences, the possibility of this reservation will be difficult to implement, as each author retains the rights to their own work.

The second provision does not provide for such reservations, cannot be overridden by contractual terms, and allows circumvention of technological protection measures to access the material (although the right must be asserted in court). In this way, the provision is more far-reaching than the first one. However, only research organisations and cultural heritage institutions have the right to copy works under it. The Swedish implementation allows for certain cases of public-private partnerships and for companies to collaborate with higher education institutions for AI research. However, companies may only form a small part of the project and have no rights to the results, which is likely to make this option unattractive to many companies.

One controversial issue is the relationship between TDM and AI. Here, the DSM Directive makes clear that TDM covers all methods, with a non-exhaustive list of examples. In response to a direct question from the European Parliament, European Commissioner Thierry Breton has replied that TDM is a method for training AI, and that TDM is the main copyright regulation in the EU for developing AI. This is made even clearer by the fact that the AI Regulation directly refers to the TDM provisions of the DSM Directive. It is thus clear that AI, including generative AI, can be trained under the TDM provisions, with the limitations described above.

The following discussion will focus on the TDM limitation for research purposes, as the possibilities for reservations, and in particular the general reservations of collective management organisations, risk rendering Section 15 a the Swedish Copyright Act ineffective. If all authors express reservations against use, AI will deteriorate and disseminate more consciously or unconsciously inaccurate information, information with bias, or poor answers to questions. Commercial AI developers will then have to rely on the riskier provision in Section 11 a, pending Swedish and European case law. To that extent, national and international organisations that publish with open access and under open licences are essential to promote positive AI development and counteract bias and disinformation. This applies to both public and non-profit organisations.

Who can make a copy?

Section 15 b of the Swedish Copyright Act gives research organisations and cultural heritage institutions the right to copy works to which they have lawful access, for the purposes of TDM. Lawful access means works to which one has legal access – via a limitation or other legal provision – or to which one has acquired permission, e.g. via a licence. Recital 14 of the DSM Directive clarifies that lawful access includes material that is legally available on the internet (such as via web crawling as discussed above, not infrequently from Wikipedia). However, the Swedish government states in its bill that "the limitation does not in itself create legal access, but rather presupposes that such access already exists, for example as a result of contracts." This means that one must have legal access through another limitation (such as the limitation on cultural heritage institutions to digitise material for research purposes in Section 16 of the Swedish Copyright Act) or through permission.

The provision focuses on institutions and organisations rather than researchers. It is thus not a limitation on researchers, but rather on institutions and organisations such as universities, colleges, libraries and museums. The Swedish government department report Ds 2021:30 states that the limitation should generally be intended to "cover employees and contractors acting on behalf of the institution or organisation or within the framework of their activities". It thereby expresses the requirement for a far-reaching and close link to the organisation or activity.

The report also notes that persons associated with research organisations and cultural heritage institutions should be covered by the limitation if, for example, they are covered by subscriptions taken out by the organisation or institution. However, it is noted that it is difficult to draw any firm conclusions about people with looser ties, or people who "act to some extent on their own behalf" (without further specification), even if they are linked to the institution or organisation. Again, this reflects a requirement for a close link to the institution or organisation, which may discourage many with shorter contracts, looser forms of employment or smaller research projects. In other words, the focus is not on the researcher and their research. The limitations also make it clear that TDM under this provision is not available to journalism, civil society or other actors important for democracy, unless it is initiated and mainly funded by public actors.

Who gets access to the copy?

The only use the limitation allows is copying for TDM training. This means that the provision says nothing about what you may, or may not, do with the copies you have made. Sweden, unlike many other countries, does not have a broad limitation on use in scientific research (based on 5(3)(a) of InfoSoc). This means that there is no other limitation in Swedish copyright law either that opens up communication of research as part of the research process, such as to other researchers, for example.

In many cases, a research group should not constitute the public. If a research group of a few people, working at the same university (the same university that produces the copies), wants to engage in TDM, sharing the copies within this small group should not be problematic (although this is not guaranteed, making it an uncertainty). However, many research projects do not look like this. EU research strategies reward large, cross-border research projects. Swedish and European funders want to promote interdisciplinary research projects. In addition, TDM requires competences that are not restricted to researchers. Most TDM projects need developers, who are not necessarily PhD researchers, and if you are working with specific tools, you may need to bring in experts in those tools, from commercial companies, academia, cultural institutions or civil society.

The limitation also discourages citizen science, for example to participate in annotating datasets. This is despite the fact that, for example, the national guidelines for open science are clear about the great value of "public participation in the research process". Manual annotation is time consuming and requires specific competences. It would be of great value if citizens could contribute to such cleaning up and thereby more qualitative AI research. The Wikimedia movement has many volunteers who are experts in categorising and structuring knowledge; whether they can also be asked to contribute their expertise to AI development when protected material is included in the training is unclear. In summary, large TDM projects require many competences, often in different universities, sometimes also outside the academic world, and often in several EU countries (especially if there is a need to understand the law in different EU countries).

The precise boundaries of what constitutes the public are not defined, but it is not at all impossible that in some research projects, especially if they last for many years and researchers come and go in the project, the research team could be classified as the public. In that case, there would be no legal basis to share the copies within the research team. Even more so if external experts and citizens are involved in the research projects.

In this context, it is also important to point out that TDM is not just a button to press. In TDM projects, large amounts of time are spent processing the works. First, works are compiled into a corpus, which then needs to be enriched, annotated or broken down in some way, depending on the type of tool and the focus of the research. For example, if text strings are to be analysed, the material needs to be processed to retain the material that is useful. At some point in this process, many of the copies are likely to cease to be works – and thus protected by copyright. At the same time, if not earlier, the use certainly ceases to infringe the economic rights of the author. However, it is far from clear where these boundaries lie. Statistical methods of analysis can be used to trace text strings and word occurrences back to the original works to a large extent. Are the text strings and word occurrences still protected? If so, without a better legal basis for making the material available within research teams, it would be virtually impossible to work with the research data within research projects.

What can you do with the copy?

The report notes that the provision allows the digitisation of works in order to carry out TDM as well as the copying necessary to structure the material and make it searchable. Accordingly, it does not only permit the specific TDM training (which might otherwise also be covered by Section 11 a of the Swedish Copyright Act). On the other hand, "research purposes" excludes, for example, the gathering of information, early experiments or studies for a potential future research project. There is simply no legal basis in the provision for investigating the possibilities for future research. The Swedish government bill also states that "there is no possibility to [...] disregard the requirement that the research must be scientific." The lawyer Tobias Kempas concludes from this that general searches for information, predictive market analyses (commercial research), "activities that cannot reasonably result in more than what already follows from proven experience" and the application of previous research results are probably not covered.

Just as there is no clear legal basis in the Swedish TDM provision for research to be shared with other researchers, there is also no legal basis for sharing it with the person who will verify it, whether for peer review or for further research – or indeed journalists or the wider public – to scrutinise the results. This is a major problem for the entire scientific process and risks undermining key research principles, as well as the objectives of transparency in national guidelines. Verifying research also requires the ability to scrutinise the underlying data. If there is no legal basis for sharing data for verification purposes, then standard scientific processes cannot be followed, in which case the legislation works against the academic process and freedom.

Since, according to many interpretations, the material may not leave the institution, it is also unclear to what extent remote access to the material can be obtained. Swedish government report SOU 2024:4 notes that an institution may only be allowed to make copies for the use of its own institution, and that in several cases researchers have had to digitise material themselves and then destroy it, even though it has already been digitised at other institutions. Legal ambiguity is a major obstacle to effective research. As discussed in the previous section of this report, a major problem is the inability to access research materials remotely, which requires research infrastructures to be built at each institution, which in turn is expensive, resource-inefficient and environmentally unfriendly. Remote access to materials is of paramount importance to qualitative TDM research, as in most cases it requires large amounts of qualitative materials. Similarly, many agreements and licences prevent researchers from accessing databases at other institutions, a waste of resources and a further barrier for researchers, discouraging efficient and qualitative research.

Where can the copy be stored and how long can it be kept

The DSM Directive is clear that copies may be kept for as long as necessary, including for verification purposes. However, this is not how it is formulated in Swedish law. While under Article 4(2) of the DSM works may be "kept as long as necessary" for TDM purposes, the Swedish implementation of Articles 3 and 4 (Sections 15 a and b of the Swedish Copyright Act) has turned the wording into a negative sentence: the copies "may not be kept longer" than required for TDM purposes (and not for other purposes). This wording also applies to the new limitation in Section 15 d proposed in SOU 2024:4. The difference seems inconsistent with EU law: the DSM Directive's wording suggests that there is no necessary end date for how long to keep the material, but that the material can be kept as long as necessary. The Swedish wording, on the contrary, suggests that there is an end date when the material must be destroyed, which puts more pressure on users to destroy material. However, for verification, scrutiny of the scientific results and reproducibility, it is crucial that material can be retained for as long as it is needed, which can be a very long time.

The negated Swedish formulation also appears to be incompatible with the presumption in the Swedish Archives Act that public documents (including research documents) must be preserved, as well as with the regulations of the National Archives of Sweden on documents from the research activities of government agencies (RA-FS 1999:1). The latter stipulates that documents deemed to be of future value for research must be preserved – and that the agency itself must decide on the application (Section 8). The negated formulation in Section 15 d presupposes that data must be deleted. If the TDM purpose here can be interpreted as the research, it includes the entire research process and then storage for at least 10 or 17 years or preservation for the foreseeable future.

What does it mean to be part of a research team

The new proposed Section 15 d of the Swedish Copyright Act opens up the possibility of sharing protected works within a "scientific research project", and all those included in the project. This is an improvement compared to the current situation and opens up for research collaborations and solves the legal grey area created by the lack of definition of communication to the public. At the same time, it is unclear what a "scientific research project" means. According to SOU 2024:4, it should be "clearly defined" in contrast to "loosely composed" research projects but is otherwise not specified further. It is also unclear what it means to be included in a research team.

According to the statute commentary, a scientific research project must be "clearly defined" as opposed to "loosely composed" research projects, but to what extent does someone have to be "included" in the project for it to be clearly defined? In order for a researcher to share material from a cultural heritage institution within a research project, they must be part of the project. At the same time, many cultural heritage institutions do not conduct their own research but do have material relevant to research. To what extent must, for example, a municipal or local history museum be included in the research project in order for researchers to be able to use their material under the limitation? Is it enough that the National Library of Sweden, as a key part of national research infrastructure, digitises and makes material available for the project for them to be counted as "included" in the project? Or must they actively participate in the research? Or is it enough that a research plan or application states that the material to be used is located at a particular institution?

Otherwise, it risks leading to the same problem as the question of who gets access to the copy. The strict limitations of Section 15 b discourage many public-private partnership projects and make exploitation through commercialisation – a priority objective of the Swedish government – impossible. For many companies, it is simply not interesting to be part of a research project if they cannot be a major part of the project or have priority access to the research results. Another major problem is that the limitations risk making citizen science impossible. It would be very problematic if citizens with expert knowledge could not, for example, participate in annotation, which is a prerequisite for successful TDM research, both for the anchoring of the research in society and for the researchers' ability to conduct effective research.

There are two TDM provisions in the Swedish Copyright Act, one for research organisations and cultural heritage institutions and one for all others. The second provision is sometimes seen as commercial. However, it applies to not only commercial actors, but also all actors that are not research organisations or cultural heritage institutions, thereby including civil society, journalists and public actors. A proposed provision included in SOU 2024:4, Section 16 f of the Swedish Copyright Act, offers some possibilities for cultural heritage institutions to make works in their collections available for, among other things, private studies. But again, this is from the perspective of the institution and not the user. Although it is good to have a clear and strong provision for TDM for research purposes, it is a democratic problem that, for example, investigative journalists and non-profit organisations do not have stronger opportunities for the same purpose.

TDM, AI and the future of the open internet

A discussion about AI and copyright is also a discussion about the future of the internet, and what shape the internet will take in the future. There are many downsides to the digital sphere, but public, private and non-profit actors have also developed tools that have changed the world, with entirely new tools at the service of society. Every day, the internet is used to promote access to knowledge, and it is an essential tool for identifying new research opportunities, studying phenomena about how society works, and promoting international cooperation. Not least through various translation applications.

The free and open internet is important, but its future depends on the positive development of AI models. Good AI models depend on a large and constant flow of data as well as the constant addition of new data. The viability of the internet, and the survival of user-generated platforms, is therefore vital not only for the platforms themselves, but also for AI systems to develop well and deliver good results. Good AI models also require an internet that is open and equitable, with actors acting in the public interest and where a diversity of voices can be heard. Otherwise, AI models will only be able to speak with one type of voice, providing one type of perspective. A diversity of perspectives via AI requires a diversity of perspectives on the internet.

In this regard, the development of open-source AI should be encouraged, especially if the AI models have been developed with public funding. Open AI would contribute to even greater societal benefits, and research funders should require that software funded through them is published with open licences. It is also essential that software and algorithms used by governments and public actors are published with open source licences, so that concerned citizens, as well as civil society and the media, can scrutinise and evaluate government actions.

Conclusion

The review presented above again demonstrates the great value of a directive-like provision corresponding to Article 5(3)(a) of the InfoSoc Directive. A broad, national implementation of 5(3)(a) would clarify the ambiguity of what lawful access means. It would undoubtedly enable communication of materials between researchers for TDM, and possibly also open up for citizen science in large TDM projects. It would probably also open up remote access to material for researchers, which means both greater fairness for researchers across Sweden, but also that the national research infrastructures that hold large amounts of works contribute the infrastructure to the country's higher education institutions.

If there is no possibility to share material from the research infrastructure (or to access it remotely), institutions and research groups are likely to either build their own corpora instead or give up certain types of research areas because access to material is too poor. The effect of the former is that similar corpora are built up in several cultural heritage institutions and research organisations, which is unduly costly and a waste of resources as digitisation and storage cost a great deal of money. It may also force researchers to affiliate with other researchers/institutions just to get access to certain material, and not according to what the research project requires. In addition, these corpora risk varying in quality depending on the availability of material, which means further geographic inequality and research results of poorer quality.

Together, Wikipedia and the other Wikimedia platforms comprise one of the main sources of material for AI and TDM, and many tools and models based on TDM and AI – such as search engines – also disseminate the material further by structuring it for the user. We think this is fundamentally positive. At the same time, there are such strong limitations embedded in the TDM provisions that problematic effects arise, and perhaps the most problematic effect is that it is difficult for the Wikimedia movement, and civil society in general, to participate in and contribute to the research process. The desire to accommodate rights holders and their representatives has led to limitations that in themselves also limit democracy.

Recommendations

• If the three-step test is reproduced in national law, only the third step should be reproduced, with a clear formulation that what is to be balanced is the exclusive right of the author and the constitutional rights of the public, not least freedom of expression, freedom of the press and academic freedom.
• Open norms provide a good basis for the development of case law, where courts can interpret broad provisions in the light of technological developments. Two provisions that could function as open norms in Swedish law are the right to quote (Section 22 of the Swedish Copyright Act) and the freedom of panorama (Section 24, Paragraph 1), but this requires harmonisation with the InfoSoc Directive.
• Article 5(3)(a) of the InfoSoc Directive should be reproduced verbatim or close to the directive in the Swedish Copyright Act, in order to promote open science and open access to science and to remove the copyright barriers that currently exist for research in Sweden. It would also provide a much better basis for so-called TDM research.
• AI models funded with public money should publish the software as open source. Software and algorithms used by government agencies and public actors should also be published as open source.
• Sweden should introduce secondary publishing rights to promote open access to research, and the government should initiate consultations to promote the dissemination of non-commercial works. All in all, this would drastically increase access to source material and knowledge in society.