Can an AI agent have a bank account?

No. A bank account requires a legal person, a verified identity, and a beneficial owner who is liable for what moves through it. Customer due diligence rules in every jurisdiction that matters were written to make sure a human or a registered entity stands behind every account, and they were written that way precisely so that nothing else could. An agent is software. It cannot sign, cannot be sued, cannot be held, and cannot be the beneficial owner of anything. It will not have a bank account, and the more interesting observation is that it does not need one.

What the question is actually asking

When someone asks whether an agent can have a bank account, they are asking one of three things, and it is worth separating them because the answers differ.

1. Can the agent hold money? Yes, in two ways that already exist. It can hold a private key to a crypto wallet, in which case it holds whatever stablecoin balance is at that address and can spend it on-chain, including through protocols like x402 that let a server ask for payment inside an HTTP response. Or it can be issued a virtual card by a card program, in which case it holds a card number with whatever controls the issuer set. Both put money in the agent's hands, and both make the agent the payer in the fullest sense.

2. Can the agent pay for things? Yes, and this is the version most people want, and it does not require the agent to hold anything. An agent can carry a key to an account its owner holds, with a policy on that key: which categories it may spend in, how much per month, and a maximum cost it must state on every call. The account pays providers on whatever rail they accept. The agent never touches the money. It holds permission.

3. Can the agent be paid? This is the hard one, and it is where the bank-account question is really pointing. An agent that earns has to receive value somewhere, and today that somewhere is either a crypto address the agent controls or an account belonging to the person or company that runs it. There is no third option, and there will not be one until the law recognizes some form of agent identity, which is not imminent.

Why the account model is the right default

An agent that holds money can lose it. That is the whole risk, and it is not hypothetical. An agent with a funded key and a bug, or a funded key and a page it scraped that told it to send everything to a particular address, drains the key. The frameworks that give agents wallets know this, which is why they ship policy engines alongside the keys and tell you in the documentation to keep balances small.

An agent that holds permission cannot lose what it does not have. Its worst case is bounded by the policy before any money moves. If it exceeds the ceiling it stated, the call is refused. If it exceeds the monthly limit on its key, the call is refused. If it asks for something outside its categories, the call is refused. If it fails, nothing is charged. The mistake is still a mistake; it is no longer an expense. And every one of those refusals is recorded against the key, which is a better audit trail than any bank statement, because it includes what the agent tried and was not allowed to do.

That is the model we built. A Vaaya account is held by a person or a company. Each agent gets a key with a policy. When a call succeeds the account settles with the provider on x402, Stripe's Machine Payments Protocol, or Tempo, and the person sees one balance and one history. We do not open bank accounts, we do not issue cards, and we do not move money between accounts. What we do is let the account holder extend bounded spending power to software, which is what the bank-account question was asking for all along.

What the account holder gets that a bank would not give them

A bank account for an agent, if such a thing existed, would give the agent a balance and a statement. It would not give the agent a history of what it bought, from whom, at what price, and what it was refused, attributed per agent, because banks do not know what an API call is. It would not size a credit line from a developer's public work, because banks do not know what a commit is. A credit account built for agents does both, and the second one matters more than it looks: it means an agent's owner can give it spending power before depositing anything, on the strength of work they have already done, and the agent's record starts accumulating from its first call.

An agent does not need a bank account. It needs an owner with one, and a policy between the two. The definitions are in what is an AI agent wallet, and the account itself is described on the credit account page.

Questions

Can an AI agent open a bank account?

No. Bank accounts require a legal person, identity verification, and a beneficial owner who is liable for the account. An agent is software and cannot be any of those. What an agent can have is a key to an account its owner holds, with a policy on what the key may spend.

How does an AI agent hold money?

Three ways in practice: a crypto wallet whose key the agent controls, a virtual card issued to it by a card program, or a key to a balance or credit line its owner holds. Only the first two put money in the agent's hands. The third gives it permission, which is usually what was wanted.