Card networks back x402 payment standard despite dispute gap in AI commerce
As AI agents begin using x402 to make online purchases, the payment standard is gaining support from major card networks even though it lacks a built-in way to reverse or contest many transactions. That gap matters most for stablecoin payments, where settlement can be final even when the agent receives incorrect, incomplete or worthless goods or services.
Highlights
- Visa, Mastercard, and American Express support the x402 payment standard, which enables sub-cent, millisecond machine payments but lacks an onchain dispute resolution system.
- Research from EPFL and Zhejiang University found that all 15 tested x402 facilitators violated at least one of eight safety rules, with nearly 2% of Base settlements reverting in late 2025—amounting to 1.9 million failed payments.
- Current x402 implementations, especially on Base and Solana where 97% of transactions occur, expose sellers to unrecoverable losses due to irreversible stablecoin settlements and lack of programmable recourse, increasing commercial risk as volumes grow.
x402 governance raises dispute concerns
As first reported by CoinDesk, the Linux Foundation switched on the x402 Foundation in July, placing Visa, Mastercard and American Express among its 17 premier members, the top governance tier for the standard AI agents use to pay for things online.The protocol is designed to support fast, low-cost machine payments, allowing transactions to settle in hundreds of milliseconds for a fraction of a cent. When Coinbase published x402 in May 2025, that trade-off favored speed and micropayment scale, but it also left out a standard dispute process for onchain stablecoin payments.
That omission is significant because the protections built into traditional card systems depend on a person spotting a problem and filing a complaint. In x402, the buyer is often software, and software does not review statements or challenge charges in the same way a human cardholder does.
The protocol can run on card rails as well as stablecoins, and card-based transactions retain existing protections. But the article says 97% of observed x402 transactions run on Base and Solana, where stablecoin payments are final once settled onchain.
Security flaws highlight commercial risk
Research published in July by the Swiss Federal Institute of Technology in Lausanne, EPFL, and Zhejiang University found that all 15 tested x402 facilitators broke at least one of eight safety rules defined by the researchers. On Base, nearly 2% of settlement attempts reverted in late 2025, equal to about 1.9 million payments.The article also says all seven of Coinbase’s official reference kits for sellers lack a way to reverse what a server already does when a payment fails to settle. In one case, the Flask kit up to version 0.2.1 releases goods once payment proof is verified, before settlement confirmation arrives, creating a risk that a seller delivers but is not paid.
The broader business problem, however, is not only failed settlement. It is the prospect that payment completes correctly while delivery, service quality or data accuracy falls short, leaving no clear machine-native framework to decide liability, evidence standards or reimbursement at micropayment scale.
That leaves the major card networks in a position to help shape what a programmable recourse system might look like. The article argues that while agent payment volumes are still small enough for losses to be manageable, the industry may need to build automated dispute and remediation tools before larger-scale failures force the issue.
Our earlier coverage of Elon Musk’s pledge to reimburse users for Grok-related transaction mistakes noted that SpaceX would “make you whole” if the agent caused financial errors or exposed card details during a purchase, even though the scope was not clearly defined. We also highlighted how most major AI-agent terms largely place transaction liability on users, as real-world mishaps—from unintended orders to oversharing personal data—underscore the growing need for clearer accountability when software transacts on a user’s behalf.