How can AI systems be a force for destabilizing authorities if they remain closed, biased and accesible only via gatekeeped API? Only truly open models (not just open-weights) with open data can have a chance to counterbalance the power dynamics. Yet most training data is concentrated in a few hands.