Key Takeaways
- On June 10, 2026, Mastercard announced Agent Pay for Machines, a platform that lets AI agents and software systems pay each other automatically across cards, bank accounts and stablecoins, with more than 30 participants including Coinbase, Stripe and Adyen
- The service transplants the trust mechanisms of the card network — authentication, spending limits and guaranteed settlement — into machine-to-machine transactions, with agent permissions and credentials initially recorded on the Polygon, Solana and Base blockchains
- Visa announced its OpenAI partnership the same day, bringing the agent-payment infrastructure race between the two card networks into the open. For e-commerce and booking businesses, this creates new questions about what they can sell to machines and how they will accept agent payments
The Gateway to an Economy Where Machines Pay Each Other
The company is working with Coinbase, Stripe and other companies to build trusted payment systems for AI-driven commerce.
www.coindesk.comOn June 10, 2026, Mastercard announced Agent Pay for Machines (AP4M), a payment platform that allows AI agents and software systems to complete payments without human intervention. It spans multiple payment methods — cards, bank accounts and stablecoins — while providing authentication, spending controls and guaranteed settlement through Mastercard's global network.
The roster of participants signals the scope of the ambition. More than 30 companies joined as initial partners, spanning payment processors, crypto infrastructure providers and developer platforms, including Coinbase, Stripe, Adyen, Checkout.com, Cloudflare, Ant International and Getnet by Santander. Chief Product Officer Jorn Lambert described the platform as creating "the conditions for a superbloom of AI business models."
Signs of demand are already showing up in the numbers. Speaking to CoinDesk, Raj Dhamodharan, who leads Mastercard's blockchain and digital asset business, pointed to rising activity around HTTP 402 — the web-standard status code meaning "payment required" — where automated transactions are failing because no payment method exists.
There are already transactions happening. There are already many declines happening because there is no payment option available. That is a leading indicator in our view.Source: Raj Dhamodharan (Mastercard EVP)
How It Works: Transplanting Trust Into Machine-to-Machine Transactions
At its core, AP4M ports the trust machinery built over decades of human card payments into machine-to-machine commerce. Dhamodharan put it plainly: "These are problems that we've solved before in the B2B world and the carded world for decades." What is new is not the problem but where it applies. According to the official announcement, the platform consists of four capabilities.
The entry point is credentialing. Every agent receives credentials, and through a mechanism called Verifiable Intent, its identity and intent can be verified across ecosystems. On top of that sits permissioning: authorization rules and spending limits set by organizations are enforced programmatically. Only verified participants can transact across providers and systems, and the chain ends with guaranteed multi-rail settlement across cards, accounts and stablecoins.
The use cases in the official announcement are telling. An entrepreneur opening a flower shop instructs an AI agent to build the store's web presence, and the agent executes a chain of transactions within budget — buying a domain, hosting, images and checkout pages. A logistics agent managing a delivery route pays for freight, reserves loading-bay access, purchases cold-chain monitoring data and settles warehouse fees automatically as the shipment moves. The platform anticipates high-frequency, low-latency transaction chains, including micropayments worth fractions of a cent.
Mastercard positions AP4M as complementary to Agent Pay, introduced in 2025: where Agent Pay defines how trusted AI agents participate in payments, AP4M handles the machine-driven microtransactions that run continuously in the background. With a B2B agent-payment pilot with HSBC also underway in Singapore, this announcement extends a steady progression from human-initiated to machine-initiated payments.
Why Record Permissions on Blockchains
The most striking design choice is that agent permissions and credentials will initially be recorded on three blockchains: Polygon, Solana and Base. Choosing public chains over the card network's own centralized ledger has a structural rationale.
In machine-to-machine payments, any counterparty must be able to verify whose delegation an agent carries and how far its authority extends — from any platform. Locking permission data inside one company's database fragments verification every time a transaction crosses ecosystems. Recording it on tamper-resistant public ledgers lets every participant reference the same information, with an audit trail produced as a byproduct. The presence of crypto infrastructure firms like Aave Labs, Alchemy, Anchorage Digital and MoonPay among the initial participants follows directly from this design.
Stablecoins also take center stage at the settlement layer. Processing sub-cent transactions through the traditional fee structure of an international card network is impractical; for small, high-frequency payments, dollar-denominated tokens on blockchains are the better fit.
Ripple moved immediately on this trend. The same day, the company released the XRPL AI Starter Kit, a developer toolkit for building AI-agent payment applications on the XRP Ledger, and joined Mastercard's agentic commerce partner roster. The kit includes an MCP server exposing XRPL documentation, skills that open wallet creation and payments to AI, and support for the x402 protocol so agents can pay for API calls in XRP or the RLUSD stablecoin. The division of labor — card networks providing trust and merchant reach, blockchains providing settlement and programmability — has now materialized at the level of developer tooling.
How It Differs From the Same-Day Visa-OpenAI Partnership
In timing that is hard to call coincidental, Visa announced its partnership with OpenAI on the same June 10. The deal embeds Visa's payment capabilities into OpenAI experiences such as ChatGPT, making it easier for merchants to accept agent-initiated Visa payments. The two dominant card networks planted their agentic-payment flags on the same day, from different angles.
| Mastercard Agent Pay for Machines | Visa × OpenAI partnership | |
|---|---|---|
| Primary focus | Machine-to-machine transactions between AI agents and software | Consumer agent purchases in ChatGPT and other OpenAI experiences |
| Payment methods | Cards, bank accounts and stablecoins | Tokenized Visa credentials |
| Structure | Open ecosystem of 30+ companies | Bilateral partnership with OpenAI |
| Permissions and controls | Permissions recorded on Polygon, Solana and Base; spending limits enforced programmatically | Spending limits, merchant categories, approval settings and real-time fraud monitoring |
| Typical transactions | High-frequency microtransactions, some below one cent | Normal-sized purchases driven by human intent |
Put simply, Visa moved to lock down consumer payments for agents that shop on people's behalf, while Mastercard moved early into the background territory where machines pay each other continuously. Both companies are hedging across consumer and machine domains, however, so this is less a division of territory than the surfacing of a race to control every layer of agent payments.
Notably, just one day before this announcement, Mastercard CEO Michael Miebach had voiced "grave concerns" about identity verification and liability in agent payments. Expressing concern and building infrastructure are not contradictory. Read together, they are two wheels of the same vehicle: precisely because trust design determines adoption, Mastercard rushed out a platform built around authentication, permissions and settlement guarantees.
What E-Commerce and Booking Businesses Should Prepare
Machine-to-machine micropayments can sound remote to e-commerce and booking operators. But AP4M is laying the foundation for a world where agents carry wallets, and in that world the way things are sold changes.
The first shift is in the unit of merchandise. The official example of a logistics agent auto-purchasing warehouse fees and monitoring data maps directly onto the periphery of e-commerce and booking operations. Inventory data feeds, delivery-slot sales, dynamically carved-out reservation capacity — businesses that package offerings machines can buy through APIs gain a new revenue channel.
The second is the payment acceptance side. The transaction failures Dhamodharan described, caused by missing payment options, are lost opportunities already accumulating today. Whether a business can identify, accept and settle agent-originated transactions becomes a new criterion in choosing payment partners. With PSPs like Adyen, Stripe and Checkout.com among the initial participants, support will likely cascade down from the payment-processing layer to individual merchants.
Conclusion
Mastercard plans to expand access to Agent Pay for Machines later this year, and the near term will center on validating use cases with partners. Viewed alongside the same-day Visa-OpenAI partnership, however, the agent-payment infrastructure race has clearly moved past the wait-and-see stage. Next to transactions where a human adds items to a cart and presses the pay button, a layer of commerce where machines pay each other continuously is quietly taking shape. Which of your offerings could be sold to machines? Carrying that question while watching each player's next move is what leads to your own.
