New analysis of 70+ blockchain-AI systems reveals most "decentralized" AI still centralizes compute off-chain, using tokens mainly for coordination. Only 10% of participatory AI projects give stakeholders real control over models. The conceptual frameworks exist - from ETHOS governance to federated taxonomies - but the gap between theory and practice exposes decentralization as mostly camouflage for concentrated power.