In most cases the company using the output can be held liable, because the infringement happens through the actual reproduction, publication or commercial use of it. Under the Law on Intellectual and Artistic Works (FSEK, No. 5846), where the similarity to an existing work reaches the level of confusion the rights holder can claim damages and an order to stop — and “the AI generated it” does not by itself make the use lawful.
Whether liability can be passed to the tool provider is a matter of contract: the supplier’s IP indemnity, its representations about the lawfulness of the training data, and who holds rights in the output all have to be set out expressly. The practical risk controls are similarity and filter checks, keeping the generation records — prompts, model versions and output logs — and human review before anything is published. Together those give you both a defensible allocation of liability and the evidence to run a defence on.
Shall we apply this matter to your situation?
Tell us your specific situation in a few sentences; we'll assess it with the right team.