The dispute at a glance
In Sweden, the National Library (KB) has a government-backed mandate to develop AI language models. To train these models, KB relies on deposit copies—the legally required copies of published works that must be sent to the library. This setup has sparked a public dispute: Sveriges Författarförbund (the Swedish Authors’ Association) and several other organizations argue that authors should not be compensated when their works are used to train AI models without explicit consent or royalties.
The legal framework behind KB’s work
KB has explained that copyright law contains balancing provisions that allow certain uses in the name of research. In recent public remarks, KB’s chief legal counsel, Karin Lodin, has asserted that the law includes a mandatory exception for research purposes, which the library believes justifies its approach to training AI language models. The goal, she says, is to steward a responsible use of copyrighted material while advancing Sweden’s AI capabilities.
The Authors’ Association stance
The Författarförbundet argues that the legal picture is unclear and that KB does not have the right to train models on its members’ works—except for the portions that fall under research. The association notes its role in defending authors’ and translators’ rights in what it calls mass usage cases, where it can be difficult for a single creator to monitor how their works are utilized. They stress the need to ensure authors receive fair compensation when works are repurposed for AI training, especially in a mass-scale context.
DN, Sweden’s daily newspaper, has reported that the association opposes blanket use of members’ works for AI training without proper remuneration. The association’s position reflects broader concerns among authors about control, fair payment, and the potential for new forms of exploitation as language models grow more capable.
KB’s perspective and the risk of overreach
KB’s legal position rests on a long-standing tradition of balancing rights with public and scientific interests. The library emphasizes that some uses of published works are restricted in order to support research and development. By applying a research exemption, KB argues it can responsibly train AI models without infringing on authors’ core rights. Critics, however, say that the absence of compensation undercuts authors’ incentives to create and participate in Sweden’s cultural and linguistic ecosystem, especially as AI tools become more pervasive in education, publishing, and industry.
What counts as research and who benefits?
Defining what qualifies as research can be tricky. The debate hinges on whether model training counts as scholarly inquiry or a broader commercial or semi-commercial activity tied to AI development. The Författarförbundet maintains that mass training of AI models using works from many authors—without direct licensing—goes beyond the boundary of protected, individualized use. They insist that clear permissions or royalties should accompany training data, particularly for translations and other derivative works.
Broader implications for Swedish AI policy
The disagreement between the Authors’ Association and KB highlights a wider tension in Sweden’s AI policy. As government agencies push for domestic AI leadership, creators worry about how their rights will be protected in an era where machine learning systems draw on vast, legally protected text. The case raises questions about future licensing frameworks, the role of deposit copies in training data, and whether a new formal compromise is needed to ensure both innovation and fair compensation.
What happens next
Resolving this conflict will likely require continued dialogue among authors, librarians, policymakers, and AI developers. Possible paths include clearer legislative clarifications on research exemptions, the establishment of compensation mechanisms for training data, or negotiated licenses for certain classes of works used in AI development. Until then, the debate underscores a critical point: as Sweden develops advanced AI tools, safeguarding creators’ rights remains a central and unresolved issue.
