Levee / Learn / One-time virtual cards vs spend mandates
One-time virtual cards vs spend mandates for AI agents
Single-use cards limit the blast radius. Spend mandates decide what the agent was allowed to buy in the first place. Most teams putting agents in charge of spend will want both.
Short answer: A one-time virtual card caps how much an AI agent can lose on that card, but its rules are frozen when it is created and whoever creates it, often the agent, holds the power. A spend mandate is a human-approved, signed statement of what the agent may buy, from whom, for what purpose, up to what amount and until when. It governs which cards may be created and is checked again on every swipe at the moment of authorization. Use one-time cards to limit exposure, and a spend mandate to decide what should happen.
What a one-time virtual card does well
A one-time (single-use) virtual card is a card number created for a single purchase or a short window. Most card issuing platforms let you set an amount, a merchant or merchant category, and an expiry when you create it. For AI agents this is a big improvement over sharing a corporate card:
- Loss is capped per card. If the number leaks or the agent misbehaves, the most it can cost is the card's limit.
- Credentials don't outlive the task. Once used or expired, the number is useless.
- Reconciliation is cleaner. One card per task makes it easy to match a charge to the job that caused it.
If an agent makes a handful of purchases a month and a person creates each card by hand, one-time cards may be all you need.
Where one-time cards stop
The rules freeze at creation
A card's limits are set when it is minted. If the price changes, a currency conversion moves the amount, or the purchase turns out to be for something else, the card has no way to know. It either fits the numbers or it doesn't.
Whoever mints the card holds the power
In agent workflows the card is often created by the agent itself through an issuing API, with the limits the agent asks for. An agent that is confused or manipulated can simply create another card. The control is only as strong as the process that decides which cards may exist.
No shared budget across cards
Ten one-time cards of $500 each are $5,000 of exposure. Per-card limits don't add up to a budget, so a team still needs something that tracks spend against an overall approval.
No purpose, no step-up, no proof
A card can check an amount and a merchant code. It can't check that the purchase is the one a person approved, it can't pause and ask a human about an unusual charge, and it doesn't record who approved what. When finance, an auditor or a dispute team asks why a payment was allowed, the card has no answer.
What a spend mandate adds
A spend mandate is a human-approved, signed instruction that defines what an AI agent may buy: merchants, purpose, amount and time window. In Levee, mandates are passkey-signed by the person approving them. Two things change when a mandate sits above your cards:
- Card creation is governed. The agent can only get a card that fits an active mandate. It can't widen its own authority by minting a bigger card.
- Every swipe is still checked. When the card is charged, the issuer's real-time authorization request is evaluated against the mandate, the platform's policy and the agent's declared purpose. The result is approve, decline or challenge, the strictest answer wins, and if no decision can be made in time the payment fails closed.
Each decision is written to a tamper-evident evidence ledger with its reason, so the question "who approved this?" always has an answer. For the full flow, see what is AI agent spend authorization?
Side by side
| Question | One-time virtual card | Spend mandate |
|---|---|---|
| How much can be lost? | Capped per card | Capped by the mandate's total budget across all cards |
| When are the rules set? | At card creation, then fixed | Approved by a human, checked again on every authorization |
| Who sets the limits? | Whoever creates the card, often the agent | The human who signs the mandate |
| Does it check purpose? | No | Yes, against the agent's declared purpose |
| Can it ask a human? | No | Yes, through a challenge |
| Is there proof of approval? | Card and transaction records only | Signed mandate plus a reasoned record of each decision |
How to use both
- Keep issuing one-time or merchant-locked cards for agent purchases. They remain your hard cap per transaction.
- Put a human-approved mandate above them. Define the purpose, merchants, total budget and end date for the agent's work.
- Gate card creation on the mandate so an agent can't create cards outside it.
- Check every authorization in real time against the mandate, outside the agent, and fail closed.
- Route edge cases to a person instead of silently approving or declining.
- Keep the evidence for finance review, chargebacks and audits.
This also matters for prompt injection. A one-time card limits the amount, but an injected agent can still spend that amount on the wrong thing. With a mandate checked at authorization, the card declines purchases outside the mandate, so the damage is bounded to what a human approved. More on the limits of static controls in why card limits and MCC blocks aren't enough for AI agents.
Where Levee fits
Levee is the authorization layer for AI agent spend. Levee lets a human approve what an agent may spend, enforces it at the moment of card authorization, outside the agent, and keeps tamper-evident proof of every decision. It is not a card issuer and never holds or moves money: cards stay with your issuing provider, and Levee works alongside the limits and single-use cards you already use.
Frequently asked questions
Are one-time virtual cards enough to control AI agent spending?
They are a strong first step because they cap the loss per card. On their own they don't check why a purchase is being made, they don't share a budget across many cards, and the rules are fixed when the card is created. For agents that buy often or on someone else's behalf, pair them with a spend mandate checked at authorization.
Who should be allowed to create virtual cards for an AI agent?
Ideally not the agent alone. If the agent can mint its own cards, it effectively sets its own limits. A better pattern is that a human-approved spend mandate defines which cards may be created, for what purpose, up to what amount and until when, and card creation is checked against it.
Does a spend mandate replace my card issuer's controls?
No. Issuer limits, merchant locks and single-use cards stay in place. A spend mandate adds a decision at the moment of authorization, outside the agent, that checks the purchase against what a human approved, and keeps a record of why it was approved or declined.
What happens if a prompt-injected agent tries to buy something with a one-time card?
With a one-time card alone, the purchase can go through if it fits the card's amount and merchant settings. With a spend mandate checked at authorization, the card declines purchases outside the mandate, so the damage is bounded to what the human approved. It does not stop the injection itself.
Does Levee issue the virtual cards?
No. Levee is not a card issuer and never holds or moves money. Cards are issued by the customer's card issuing provider; Levee decides whether each agent payment should go through and keeps tamper-evident proof of every decision.
Related
Levee is running a free design partner pilot for platforms putting AI agents in charge of spend.
Become a design partner