ARTICLE
Product
Safe4 runs four checks between an agent deciding to spend and the money moving: policy, purpose, budget, and counterparty. A payment must clear all four before it executes, and any single failure stops it.
What does each check ask?
Policy
Whether this class of payment is permitted at all — the rules that hold regardless of task or context.
Purpose
Whether the payment belongs to the task the agent was actually given. A payment can be in policy and still be off purpose.
Budget
Whether it fits the envelope that task was funded with, counting everything already spent against it.
Counterparty
Whether the recipient is one this agent is allowed to pay.
Why do all four have to clear?
Because each check catches a failure the others miss. A payment to an approved vendor, within budget, under an allowed policy, is still the wrong payment if it has nothing to do with the task at hand.
Why run the checks before execution?
Reconciliation after settlement tells you what went wrong once the money has already left. A control that only reports is not a control.
The same reasoning sits behind Google’s Agent Payments Protocol (AP2), announced on 16 September 2025 with more than 60 partner organisations. AP2 frames agent payments around three questions: authorization, authenticity, and accountability — proving the user granted the authority, that the request reflects their intent, and who is answerable when it goes wrong.
Frequently asked questions
What happens when a check fails?
The payment does not execute. The decision — allow, deny, or hold — is returned to the calling agent rather than surfaced after the fact.
How is a budget check different from a spending limit?
A spending limit is a ceiling on a single transaction. A budget is an envelope attached to a task, and it accounts for everything already spent against that task, which is what catches loops and retries.