AI agents are increasingly able to initiate and complete transactions directly, on timelines too fast for a person to review each step. That changes what a trust decision needs to look like: it can no longer rely on a human noticing something is wrong at the point of action, because there may be no such point to notice it at.
What has to happen earlier
If a transaction is going to be executed autonomously, the questions a human decision-maker would normally ask — who made this claim, is the source authoritative, is the information still valid, is this authorised, is this appropriate for the purpose — have to be answerable programmatically, before the transaction executes, rather than reviewed afterwards. This is the same requirement set out in AI Trust, applied specifically to the moment of execution.
Relationship to machine and delegated authority
An autonomous transaction is only as trustworthy as the authority behind it. See Machine Authority for what the acting system needs to be able to demonstrate, and Delegated Authority for how that authority is granted and bounded in the first place.