Short answerA good agent mandate names the agent, the allowed payees, per-payment and monthly limits, the accounts, an expiry date and what happens on exceptions, and is signed by the account holder with a strong method such as a passkey. The bank's ceilings sit above it.
Example
"My agent may pay approved suppliers up to $2,000 each and $15,000 a month from Operating ••4821 until 31 March 2027. Anything new or larger, ask me."
Required fields
| Field | Why it matters |
|---|---|
| Agent | Ties the permission to one identified agent, not "any AI". |
| Payees or payee types | New payees are the most common fraud path, so they should trigger approval. |
| Per-payment limit | Caps the damage of a single wrong action. |
| Monthly limit | Caps cumulative exposure. |
| Accounts | Keeps the agent off accounts it does not need. |
| Expiry | Forces periodic review. |
| Exception rule | Says what happens outside the mandate: ask, decline or require two approvers. |
| Signature and version | Shows who agreed to which version and when. |
Bank ceilings versus customer mandates
The bank sets defaults and maximums by account type. The customer signs a mandate within them. Neither can exceed what the other allows.
Common mistakes
- Open-ended scopes like "manage my payments".
- No expiry.
- Allowing new payees without approval.
- Storing the mandate somewhere the dispute team cannot retrieve with the payment.
How mandates relate to protocols
Agent protocols handle how an agent connects and requests scopes. Google's AP2 uses signed mandates for commerce; the Personal Agent Protocol draft defines read and write scopes and lists payments and signed approvals as open work. A bank still needs to enforce the customer's limits on its own side.