Sweden’s Authors’ Union challenges KB over AI training
In Sweden, the Authors’ Union has publicly urged the National Library of Sweden (KB) to stop training its upcoming Swedish AI language models on works by the union’s members. The union argues that the legal framework is unclear and that such training could exploit writers rights without compensation. The controversy reflects a broader debate about how to balance cultural creators’ rights with the societal benefits of AI research.
What is at stake
The background is that KB, under a government assignment, aims to build AI language models by using pliktexemplar, the mandatory copies of published works that must be submitted to KB. The union warns that a broad use of these works for training, especially in mass quantities, may undermine authors control over their creations and set a precedent for future exploitation. As a result, rights holders fear diminished incentives to publish and translate if their works are repeatedly used without explicit consent or compensation.
KB’s interpretation and the research exception
KB has signaled a different interpretation of copyright law. In a prior discussion, KBs chief legal officer, Karin Lodin, argued that copyright law balances authors rights with other societal interests and that there exists a mandatory exemption for research purposes. KB contends that this provision allows their approach and that they feel secure in applying it to training data for AI. Critics, however, point to gaps in public understanding of how such exemptions apply to mass data processing and to who ultimately benefits from AI models built on these datasets.
The authors rights dimension
The union insists that mass utilisation raises questions about fair compensation and consent, particularly for rights holders who may not be aware of how their works are used in AI training. Forfattarförbundet has said its role is to protect authors and translators rights in mass uses that are difficult for an individual creator to monitor. The union argues that without transparent licensing or revenue-sharing schemes, the use of works for AI training could erode confidence in the publishing ecosystem and harm authors livelihoods over time.
The broader debate
Several author associations and industry groups have highlighted concerns that their members works could be appropriated to train AI without remuneration. The conversation in Sweden touches on ownership, dataset rights, and how to implement a credible research exception without undermining the value of creative works. Debates abroad mirror this tension, underscoring the need for clear national guidance on data use, consent, and compensation in AI development.
Next steps and possible paths forward
Observers expect a period of negotiation and potential legal clarification. Possible routes include explicit research exemptions tailored for AI training, licensing agreements with rights holders, or programmatic safeguards to ensure transparent use and fair compensation. The Authors’ Union advocates for policies that secure consent and a fair return for authors and translators, rather than relying solely on institutional benefit. A transparent, multi-stakeholder process could help align research aims with creators rights and public trust in AI technologies.
Why this matters for AI development in Sweden
How Sweden balances the needs of advancing AI with protecting creators rights will shape the future of research, culture, and innovation. A framework that respects rights while facilitating responsible AI training could encourage ongoing collaboration between authors and technologists. Conversely, persistent uncertainty risks a chilling effect, discouraging authors from sharing works and hindering potential breakthroughs in Swedish AI research.
Conclusion
The dispute between the Authors’ Union and KB highlights a central challenge of modern AI policy: enabling technological progress while upholding the rights of creators. As discussions continue, policymakers, rights holders, and research institutions will need a transparent process that respects authors rights and ensures fair returns, creating a sustainable path for AI development in Sweden.
