Notes from the health layer

AI security research has independently converged on one architectural principle: don't try to make the model trustworthy — constrain its powers and judge it by observable behaviour. That is exactly the principle we apply to on-chain agentic commerce.