An AI agent can propose a wallet transaction, but how can the wallet check that it came from the agent the user intended to authorise? Through a worked example, we outline the trust placed in the model provider and agent host, then introduce VET and the GUARD research direction for authenticating delegated actions before granting wallet access.