What Is x402 and How Do AI Agents Pay for Things?

What problem does x402 solve?

The web has always had a status code for payment. HTTP 402 Payment Required has been part of the standard for decades, but almost nobody used it. Websites asked for payment with sign-up pages, card forms and subscriptions instead.

That model does not work for AI agents. An agent that needs one weather report, one data lookup or one page of a paid article cannot open an account for each service. It also must not hold a company credit card. x402 gives the 402 code a real job. It lets a client pay for one request, in a few seconds, with money that moves on a blockchain.

Coinbase published x402 as an open protocol in 2025. Other companies now build on it. The details can change between versions, so check the current specification before you build. The core idea stays the same.

How does an x402 payment work step by step?

1. The agent sends a normal HTTP request to an API.
2. The server answers with status 402. The answer lists the price, the accepted token, the blockchain network and the address that receives the payment.
3. The agent’s wallet signs a payment for that amount, usually in a stablecoin. A stablecoin is a token that keeps a steady value, for example one US dollar.
4. The agent sends the same request again, with the signed payment in a request header.
5. The server checks the payment and settles it on-chain. Many servers use a separate service, often called a facilitator, to do the check and settlement.
6. The server returns the data, with a payment response that the agent can keep as a receipt.

The agent needs no account, no API key and no card. The server needs no billing system. Each call stands alone.

What are agentic payments in general?

Agentic payments are payments that an AI agent starts for a person or a company. x402 is one way to pay for API calls. Agents also buy goods, book travel, pay invoices and pay other agents. All of these share the same questions.

• Who authorizes the payment? A person, a company policy or the agent alone.
• What are the limits? Amount per payment, per day, per seller and per task.
• What proof stays behind? A receipt that shows who paid, what for and under which rule.

The answer to the first two questions is a mandate. A mandate is a signed instruction from a person that says what an agent can buy, from whom and up to what amount. For example, a mandate can allow data purchases from three named APIs, up to 20 US dollars per day, until the end of the month. The agent can act inside the mandate. Anything outside it needs a new approval.

What can go wrong when agents pay?

Payments turn every agent attack into a money problem. Our threat map for AI agents covers the attacks in full. These are the ones that matter most for payments.

Runaway spend. An agent that pays per call can loop. A small price times many thousands of retries becomes a large bill. The agent does not feel the cost, so it does not stop.

Fraud by fake sellers. An attacker runs an API that answers every request with 402 and a high price. Or the attacker copies a real service and changes the receiving address. The agent pays because the response looks normal.

Prompt-injected payments. A web page or file contains hidden text that tells the agent to pay 50 dollars in stablecoins to an address to finish the task. The agent reads the text and treats it as a step in the task. This is prompt injection with a wallet attached.

Price changes. A server quotes one price and then asks for a higher one on the retry. An agent with no limit pays the new price.

Lost records. Without a strong receipt, a finance team cannot tell which agent paid, for which task and under which mandate.

How do on-chain rules keep agent payments safe?

A prompt can tell an agent to be careful. It cannot stop the agent. Payment rules must sit outside the model, where injected text cannot change them.

Risk	Control	Where it runs in Pavii
Runaway spend	Budgets per agent, task and day	Budget gate
Fake sellers	Allow-list of sellers and addresses	Egress smart contract
Injected payments	Pay only for user-started tasks	Injection gate plus egress gate
Price changes	Maximum price per call	Egress smart contract
Large payments	Human approval above a limit	Gateway approval step
Lost records	Hash-chained receipts	Receipts anchored on Hyper 0X

In Pavii, a payment is a tool call like any other, and all six gates check it. The budget gate stops spend at the limit. The egress smart contract on Hyper 0X decides which addresses an agent can pay and how much. A smart contract is a program on a blockchain that runs the same way every time. The rule still applies when a long task fills the model’s context window and the model forgets an instruction.

If an agent breaks a payment rule, the kill switch moves it from observed to restricted to stopped. A person can stop any agent at any time. For the full design, read how blockchain is used to secure AI and our prevention playbook.

Who is responsible when an agent pays the wrong party?

The company that runs the agent is responsible in most cases. So keep proof that the agent acted inside its mandate. Each Pavii receipt holds the agent identity, the rule that allowed the payment and the gate results. Pavii links receipts in a hash chain and anchors them on Hyper 0X, so nobody can secretly change or delete a record.

Good receipts also help in disputes with sellers. You can show the price that the server quoted, the amount the agent paid and the time of each step. Buckleson builds the chain behind this. See Pavii on buckleson.com.

Where should you start?

1. List every place where an agent can spend money today, including API keys with billing.
2. Write a mandate for each agent: sellers, limits per call and per day, and an end date.
3. Give each paying agent its own wallet with a small balance.
4. Set human approval for any payment above your chosen limit.
5. Put all payment calls behind a gate that checks budgets and allow-lists, such as Pavii Security.
6. Keep a receipt for every payment and review them each week.

Book a 30-minute call with the PAVii.AI team to design safe payments for your agents.