European media organisations are no longer passive users of AI. Their archives, editorial expertise and audience relationships have become valuable inputs to AI training, retrieval and product development. Yet contributing does not automatically translate into control, revenue or strategic autonomy.

Whoever holds the archive does not automatically control what gets built from it. That is a problem underneath most current debates about digital sovereignty, and it lands hardest on organisations with valuable holdings and little bargaining power. Broadcasters, publishers and heritage archives are supplying provenance-rich, rights-encumbered material into systems they neither own nor govern. Large players negotiate bilateral deals under confidential terms. Smaller ones cannot negotiate at all.

This session presents work from VLAM (Vlaams AI Modellen), an imec-led programme building multimodal AI for Flemish use cases, in collaboration with, among others, VRT, DPG Media, and Roularta. The programme asks how Flanders and Flemish companies can decide for themselves whether, how, and under which conditions their own material is deployed. Not by training foundation models, but by adapting existing ones to the Flemish language, content, and context.

This talk discusses a mapping of the European media sector's position across compute, models and applications, and ongoing work on how such a programme should be governed and how its outputs should be valorised.