x402 Adds Card Networks to Governance, but AI Agent Payments Still Lack Chargeback Rules

x402 Adds Card Networks to Governance, but AI Agent Payments Still Lack Chargeback Rules

N
News Editor
2026-10-29 02:03:51
The Linux Foundation launched the x402 Foundation in July, putting governance of the AI-agent payment standard under a new structure whose top tier includes Visa, Mastercard, and American Express. The setup has drawn attention because x402 was designed without a standard dispute-resolution layer: once a payment settles on-chain, there may be no issuer, no chargeback path, and no clear way to challenge a bad outcome. That gap matters as AI agents begin buying data, calling APIs, and renting compute on their own, often in transactions worth only a few cents or less. A recent column by GenLayer Labs co-founder Albert Castellana highlighted the contrast between card networks’ long history of dispute handling and their presence in the governance of a system that does not yet offer an equivalent mechanism. A July study from EPFL and Zhejiang University tested 15 x402 facilitators and found that every one of them violated at least one of eight security rules defined by the researchers. The paper also said roughly 2% of settlement attempts on Base had failed and rolled back by the end of 2025, equal to about 1.9 million payments that did not actually complete. On the merchant side, Coinbase’s seven official seller examples were also described as lacking a way to roll back actions after failed settlement.

As AI agents start buying data online, calling APIs, and renting compute on their own, a transaction can finish in a few hundred milliseconds and a wallet balance can drop by $0.03. But if the data returned is wrong, there may be no familiar dispute path to follow. That is the issue raised around x402, a payment standard built for AI agents.

In July, the Linux Foundation said the x402 Foundation had officially launched to take over governance of the standard. Among the 17 members at the top governance layer are Visa, Mastercard, and American Express.

Albert Castellana, co-founder of GenLayer Labs, described the arrangement bluntly in a recent institutional newsletter column: some of the world’s largest card networks now sit at the highest governance level of a payment standard that does not have a standard dispute-resolution process.

x402 leaves out the dispute layer on purpose

At the launch, Mastercard said native internet payments need to be as safe, trusted, and interoperable as digital commerce is today.

In the card world, that promise rests on a long-established framework. Since the U.S. Fair Credit Billing Act was passed in 1974, consumers have had the right to dispute credit card charges. If goods do not arrive or services are not delivered, cardholders can go through the issuing bank to try to get their money back.

x402 removes that layer, and it does so by design. On-chain settlement can be pushed down to a few hundred milliseconds, with per-transaction costs below one cent. That makes machine-scale micropayments possible in a way card networks are not structurally built to handle. When Coinbase unveiled the protocol in May 2025, on-chain agent payment volume was still thin enough to look experimental, so prioritizing speed first and leaving disputes for later was presented as a reasonable trade-off at the time.

The deeper difference is the identity of the buyer. The defenses built into the card system over the past 50 years assume that a person will notice something is wrong and call to complain. In x402, the buyer is often software. Software does not read statements, and it does not remember three months later that a $0.03 charge looked odd.

Study finds all 15 facilitators broke at least one security rule

In July, a research team from the Swiss Federal Institute of Technology Lausanne (EPFL) and Zhejiang University published what it described as the first systematic security study of x402 deployments running online. The team tested 15 facilitators, the intermediaries responsible for validating payments and pushing transactions on-chain.

Every one of the 15 tested facilitators violated at least one of the eight security rules defined in the study. On Base, about 2% of settlement attempts had failed and rolled back by the end of 2025, which the researchers translated into roughly 1.9 million payments that did not actually complete.

The study also said 97% of observed x402 transactions ran on chains with finality.

Merchant-side reference implementations also fell short

The merchant side showed similar gaps. Coinbase’s seven official seller examples all lacked the ability to reverse actions already executed by a server when payment failed to settle successfully.

One example stood out. Before version 0.2.1, the Flask sample released goods after proof of payment had been verified but before settlement had been confirmed. If settlement later failed, the seller had already delivered while receiving no funds.

Early infrastructure bugs are not unusual on their own. The harder problem is that payment, delivery, and settlement can each tell a different story. The buyer may think payment was made, the seller may think funds were received, and the chain may show the case as closed. If one link breaks, no participant is necessarily required to stitch the process back together.

Final payment does not settle whether the deal was fulfilled

The reverse scenario is even harder: payment lands exactly as intended, but the agent receives something wrong, incomplete, or useless. x402 supports a card rail as well, and transactions routed through cards still keep card protections. But 97% of observed x402 transactions ran on Base and Solana, where stablecoin payments are generally irreversible once confirmed.

Payment finality is not the same thing as commercial finality. A blockchain can prove that money moved. It cannot decide whether the data was correct, whether a service was delivered, or whether both sides performed as promised.

The card industry spent 50 years building answers to dispute questions, and each answer came with a cost allocation: who absorbs the loss, who decides whether service was delivered, how long the buyer has to complain, and what counts as evidence. When a transaction is worth one-tenth of a cent, those questions still do not have settled answers.

What to watch next

The first point to watch is governance. The top layer of the x402 Foundation holds some of the deepest dispute-management knowledge in payments, including how much a dispute process costs and how much trust it buys. Whether that knowledge can be rewritten into a machine-native form may determine whether x402 becomes commercial infrastructure or remains an experiment.

The second is scale. Agent payments are still small enough that a bad month may be manageable. If financial institutions connect tokenized assets to agent payments, the cost of a settlement failure will no longer be measured in a few cents.

Over the past few years, the industry has made money programmable. The next question is whether recourse can be made programmable too. That is not framed here as a pure technical barrier. It is a design choice: build it before losses force the issue, or wait until the first large-scale accounting failure and patch it later.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
7900

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.

x402 Adds Card Networks to Governance, but AI Agent Payments Still Lack Chargeback Rules | Bit.Fan