What to Know
- Visa, Mastercard and American Express are among the 17 premier members of the x402 Foundation, the top governance tier for the AI agent payment standard.
- The Linux Foundation switched on the x402 Foundation in July, after Coinbase published the protocol in May 2025.
- x402 is designed to let AI agents pay for online goods and services, often through stablecoin payments that can settle in hundreds of milliseconds and cost a fraction of a cent.
- The standard has no built-in, standardized dispute process for stablecoin transactions once settlement occurs onchain.
- U.S. credit card users have had a legal right to dispute charges since the Fair Credit Billing Act passed in 1974.
- Researchers from EPFL and Zhejiang University tested 15 x402 facilitators and found that every one broke at least one of eight safety rules they defined.
- On Base, close to 2% of settlement attempts reverted in late 2025, equal to around 1.9 million payments.
- All seven of Coinbase’s official reference kits for sellers lacked a way to undo what a server already did when payment failed to settle.
- About 97% of observed x402 transactions run on Base and Solana, where stablecoin payments are final.
- The central unresolved issue is commercial recourse: who absorbs the loss when a payment settles but the agent receives something wrong, incomplete or useless?
Card Giants Step Into AI Agent Payments
The rapid development of AI agent commerce is pushing one of digital payments’ oldest problems into a new, machine-driven environment. The x402 standard, built to let software agents pay for goods and services online, now has the backing of the world’s largest card networks at the governance level. Visa, Mastercard and American Express are among the 17 premier members of the x402 Foundation, the top tier overseeing the standard under the Linux Foundation.
The move signals that traditional payments leaders see internet-native, automated payments as a serious infrastructure category. It also puts them close to a difficult design choice. The x402 model can make payments fast, cheap and programmable, particularly when stablecoins are used. Yet stablecoin settlement on public blockchains does not come with the same consumer protection architecture that card users have relied on for decades.
That contrast matters because AI agents are expected to transact without direct human involvement in every purchase. A person can notice an unfamiliar credit card charge, complain to a bank and begin a dispute. Software can execute instructions, verify a payment format and consume a service, but it does not naturally review a monthly statement or understand that the delivered result was commercially worthless unless it has been specifically designed to detect that failure.
The Chargeback Problem Moves Onchain
Card payments are not trusted only because they move money. They are trusted because they sit inside a broader process for reversing mistakes, investigating fraud and allocating losses. In the United States, the right to dispute a credit card charge has been embedded in law since the Fair Credit Billing Act passed in 1974. That recourse helped people become comfortable buying from unfamiliar sellers, including on the internet.
x402 takes a different approach. For stablecoin payments, once funds settle onchain, there is no card issuer to call and no standard network chargeback to file. That finality is a feature for some use cases. It reduces friction, makes settlement predictable and allows extremely small machine-scale payments that would be impractical under many legacy payment systems. But payment finality is not the same thing as commercial finality.
A blockchain can show that value moved from one address to another. It cannot, by itself, decide whether a data feed was accurate, whether an API response was useful, whether a file was complete or whether a digital service matched what the buyer requested. That gap becomes more significant when the buyer is an AI agent making frequent, low-value payments at speed.
Market participants watching x402’s development say the missing dispute layer could become more important as usage grows. At small volumes, errors may be treated as an operational inconvenience. At scale, even tiny transaction values can accumulate into meaningful losses if agents regularly pay for wrong, incomplete or unusable output.
Why x402 Left Disputes Out
The absence of a standard chargeback process is not necessarily an oversight. It reflects a trade-off at the heart of the protocol’s design. Coinbase published x402 in May 2025, when usage remained sparse and the priority was enabling a new class of internet-native payments. The promise was clear: payments that can settle in hundreds of milliseconds at a cost of a fraction of a cent.
Those economics are important for AI agents. A software agent might need to pay for a single API call, a small data query, a brief compute task or a small online resource. Traditional dispute systems were built for human commerce, where the administrative cost of reviewing a claim can be justified by the size of the transaction. In machine commerce, many payments may be too small to support human review.
That is why the next phase of x402 may require a dispute model that looks different from traditional card chargebacks. Human-driven claims, long review windows and manual evidence gathering may not fit transactions measured in fractions of a cent. Still, the commercial need remains: buyers and sellers need rules for what happens when payment and delivery do not align.
Security Research Highlights Early Risks
Research published in July by the Swiss Federal Institute of Technology in Lausanne and Zhejiang University underscored the operational challenges already visible in live x402 deployments. The researchers tested 15 facilitators, the intermediaries responsible for verifying payments and pushing them onchain. Every facilitator broke at least one of eight safety rules defined in the study.
The issue was not only theoretical. On Base, close to 2% of settlement attempts reverted in late 2025, representing around 1.9 million payments. A reverted settlement can create a mismatch between what a seller’s server believes happened and what actually settled onchain. If goods are released before settlement is confirmed, the seller may deliver without receiving payment.
One example involved Coinbase’s official reference kits for sellers. All seven lacked a way to undo what a server had already done when payment failed to settle. In the Flask kit up to version 0.2.1, goods could be released once a payment proof verified, before settlement was confirmed. If settlement later failed, the seller had already delivered and would not be paid.
Early infrastructure frequently contains bugs, especially in new markets. The larger lesson is that payment, delivery and settlement can diverge. That creates two types of risk. In one case, the seller delivers and is not paid. In the other, the buyer pays and receives something wrong, incomplete or useless. The second scenario is especially important for AI agents because payment may settle perfectly even while the commercial outcome fails.
Stablecoins Dominate Observed x402 Activity
The x402 protocol can support card rails as well as stablecoins. When card rails are used, card protections can still apply. However, observed activity has been overwhelmingly centered on blockchain-based stablecoin payments. About 97% of observed x402 transactions run on Base and Solana, where stablecoin payments are final.
That means the practical dispute question is largely an onchain question today. Stablecoins are useful for programmable settlement because they can move across blockchain infrastructure without relying on a conventional card authorization and chargeback process. But that same design removes the familiar safety net for many participants.
For institutions, the governance role of major card networks is therefore notable. These companies have spent decades building rulebooks around disputes, fraud, merchant obligations, evidence standards and loss allocation. Their involvement does not automatically solve x402’s recourse gap, but it places deep payments expertise near the standard’s evolution.
The Commercial Question: Who Covers the Loss?
The hardest questions are not purely technical. If an AI agent pays for data that is stale, a service that is incomplete or content that does not match the request, who decides whether the buyer has a valid complaint? How long should the buyer have to object? What evidence should a seller provide? Can the process be automated? Who absorbs the cost when the payment amount is too small to justify human intervention?
Traditional card systems answered versions of those questions over more than five decades, but those answers were expensive. They required issuers, acquirers, networks, merchants, procedures, documentation and arbitration. Machine-scale commerce may need a lighter, faster and more programmable form of recourse.
Some chart watchers and infrastructure builders frame the issue as the next stage of programmable money. Blockchains made settlement programmable. Stablecoins made value transfer more internet-native. AI agents may now require programmable recourse: automated methods for escrow, conditional release, reputation, verification, refunds or dispute routing when the delivered result fails to meet expectations.
No single mechanism has emerged as the standard answer. Escrow can protect buyers but may add friction. Reputation can guide agent behavior but may not compensate a loss. Cryptographic proofs can confirm that something was delivered, but not always that it was useful. Insurance and pooled risk models may help in some cases, but they introduce pricing and governance questions.
Why Institutions Are Watching
The x402 debate comes as broader institutional crypto activity continues to expand across tokenization, stablecoins and regulated digital asset products. Banks and asset managers are experimenting with tokenized money, deposit tokens, tokenized investments and bitcoin-linked funds, while regulators draw boundaries around what these products can offer.
In that environment, AI agent payments occupy a strategically important niche. They connect stablecoins, public blockchains, automated software and online commerce. If x402 or similar standards become widely used, they could shape how machines pay for data, compute, software services and digital goods.
But institutional adoption typically requires more than fast settlement. It also requires operational controls, legal clarity and predictable loss allocation. The involvement of major card networks suggests that the payments industry understands the opportunity. The unresolved question is whether the industry can build a machine-native protection model before larger losses force the issue.
Programmable Money Needs Programmable Recourse
The core issue is not whether x402 is useful. Its design addresses a real need: low-cost, fast, automated payments for internet-native services. The challenge is that commerce is broader than settlement. A complete payment system must handle failure, disagreement and fraud, especially when buyers and sellers do not know each other.
For now, agent payment volumes remain small enough that a difficult period may be manageable. But the direction of travel is clear. If AI agents become active economic participants online, they will need ways to resolve mistakes without relying on a human to notice every problem. That makes dispute design one of the most important open questions in AI-driven crypto payments.
FXCOINZ views the x402 debate as a sign that digital asset infrastructure is moving beyond simple transfer mechanics. The next competitive frontier may be trust. Fast final settlement can move value efficiently, but market participants still need confidence that a payment system can respond when the transaction succeeds technically and fails commercially.
Frequently Asked Questions (FAQs)
What is x402?
x402 is a payment standard designed to let AI agents pay for goods and services online, including through stablecoin-based payments that can settle quickly and at very low cost.
Who governs the x402 Foundation?
The x402 Foundation operates under the Linux Foundation, and its top governance tier includes 17 premier members. Visa, Mastercard and American Express are among those members.
Why is the dispute issue important?
The issue matters because stablecoin payments that settle onchain are generally final. If an AI agent pays and receives something wrong, incomplete or useless, there is no standard x402 chargeback process for stablecoin transactions.
How is this different from credit cards?
Credit card systems include dispute and chargeback mechanisms. In the United States, the right to dispute a credit card charge has existed since the Fair Credit Billing Act passed in 1974.
Does x402 support card payments?
Yes. The protocol can support card rails as well as stablecoins, and card payments can retain card-based protections. However, about 97% of observed x402 transactions run on Base and Solana, where stablecoin payments are final.
What did researchers find about x402 deployments?
Researchers from EPFL and Zhejiang University tested 15 facilitators and found that every one broke at least one of eight safety rules they defined. On Base, close to 2% of settlement attempts reverted in late 2025, equal to around 1.9 million payments.
What is a facilitator in x402?
A facilitator is an intermediary that verifies a payment and pushes it onchain. Facilitators play an important role in connecting payment proof, settlement and delivery.
Why can payment and delivery disagree?
Payment and delivery can disagree when a system treats a payment proof as sufficient before settlement is confirmed, or when settlement succeeds but the delivered service or data does not meet the buyer’s needs.
What might a solution look like?
Potential approaches could include automated escrow, conditional release, reputation systems, refunds, verification tools or machine-readable dispute rules. Any solution would need to work at the speed and scale of agent payments.
