Who actually reads the AI Act's GPAI transparency rules?

The transparency obligations for general-purpose models are now in force, but I keep meeting developers who assume "open weights" means "out of scope." It doesn't, and the paperwork burden for small teams looks brutal. The training-data summary requirement especially seems vague enough that everyone will interpret it differently. I'm not against the rules, but compliance guidance from national authorities is thin on the ground. Has anyone here actually filed a model card or summary that satisfied a regulator? Curious whether this is genuinely manageable for a five-person lab or whether it quietly consolidates power with the big providers.