Agent mandates

What is an agent mandate?

An agent mandate is a machine-verifiable, legally worded declaration by a legal person that a specific AI agent may perform defined actions on its behalf — limited by action, amount, frequency, counterparty and time, revocable at any moment, provable for both sides.

The kill switch: revocation as the principal's basic right

The kill switch is the non-negotiable element of every agent mandate: the person granting the mandate can end any authority at any time, without giving reasons. Revocation is atomic — status change, evidence entry and webhook commit together — and takes effect from the next verification, without a cache.

Escalation: when a human has to decide

Escalation (step-up) is the mechanism that reconciles autonomy and oversight: below the threshold the agent acts alone; above it the person who granted the mandate is asked — with context, a timeout (600 seconds) and deny as the default.

Four-eyes principle: organisational mandates with signing rules

Organisations grant authority differently from individuals: above a defined threshold an organisational mandate needs two signatures from authorised signatories — and that rule must be enforced, not recommended.

Constraints: amount, frequency, geography, time windows

Constraints turn a blanket permission into precise authority: amounts limited per action and per period (reserved atomically against race conditions), frequency limits per day or month, geographic allow-lists by ISO 3166 and UTC time windows.