EU AI Act: Are open-source models at risk?

I've been reading the final text of the AI Act and there's a lot of talk about exemptions for open-source. But the more I dig in, the more I wonder if the transparency and documentation requirements effectively kill small projects. The big labs have legal teams to handle everything, but a hobbyist or a small startup? Releasing a model's training data details or providing that technical documentation is not trivial. Are we just creating a gatekeeper environment under the guise of safety? I'm genuinely curious what others think. Is the exemption window wide enough, or is this an accidental regulatory barrier that only the big players can afford to navigate?