Mastercard’s AI Payment Roadmap: From Agent Pay to Machines Paying Machines

Mastercard is positioning itself for a payments market in which software does more than recommend a purchase: it can authenticate, authorize, and complete the transaction itself. CEO Michael Miebach’s roadmap starts with verified AI agents acting for consumers and extends to machines paying other machines continuously, sometimes for very small amounts.

The strategy is less about replacing the card network than about making Mastercard the trust and settlement layer behind new kinds of commerce. Stablecoins fit into that plan as another payment rail, particularly for dollar-based and cross-border transactions—not as a replacement for the dollar.

The roadmap has two distinct stages

1. Agent Pay gives AI agents an accountable identity

Mastercard introduced Agent Pay in April 2025 to address a problem that does not exist in ordinary card payments: the software submitting the transaction may not be the consumer or the merchant. Mastercard describes the agent as an additional participant that must be registered and verified before it can transact on a user’s behalf.

The system uses Agentic Tokens, which extend Mastercard’s existing tokenization approach. Rather than giving an AI agent a reusable card number, the token can connect the payment credential to the specific agent, merchant scope, and consent rules governing the transaction.

That distinction is important. A conventional card credential answers, broadly, “which account should be charged?” An agentic credential also needs to help answer:

• Which software agent is acting?
• Which person or organization authorized it?
• What is the agent allowed to buy?
• Are the transaction and its context still within those permissions?

Mastercard’s framework therefore treats agent registration, verification, permissioning, and traceability as core payment functions rather than optional add-ons.

2. AP4M extends the model to machine-to-machine commerce

In June 2026, Mastercard introduced Agent Pay for Machines, or AP4M. The service builds on Agent Pay but targets a different use case: autonomous agents and connected systems making high-frequency, low-latency, low-value payments without a person approving every individual transaction.

Potential transactions include an AI agent purchasing access to data, an API, or computing capacity as needed. Instead of arranging one large subscription or requiring a human checkout, an agent could pay for resources incrementally at machine speed. The launch materials describe support for payments across cards, bank accounts, and stablecoins, including transactions worth fractions of a cent.

AP4M’s key components are not simply faster authorization. They include:

• Credentialing: identifying agents so counterparties can distinguish trusted software from an unknown bot.
• Permissioning: applying spending rules, limits, and other controls set by the relevant organization or user.
• Orchestration: coordinating the interaction between the paying and receiving systems.
• Settlement: completing payment across more than one underlying rail.

Mastercard’s stated goal is to provide a common network layer for those functions, rather than requiring every AI platform, merchant, and infrastructure provider to build a separate closed system.

Why the partner network matters

AP4M launched with more than 30 participants and supporters from payments, crypto infrastructure, cloud services, and blockchain networks. The named ecosystem includes companies and organizations such as Stripe, Coinbase, Ripple, the Solana Foundation, Polygon, and Cloudflare.

That mix reflects the problem Mastercard is trying to solve. Machine-to-machine commerce requires more than a payment processor. It needs:

1. Agents that can prove who they are.
2. Owners that can define and revoke permissions.
3. Merchants and service providers that can verify intent.
4. Payment rails that can handle both traditional and tokenized money.
5. Settlement systems that work across different ecosystems.

The more participants support the same trust and payment conventions, the less likely the market is to fragment into incompatible agent economies. But the number of launch partners should not be confused with proof of mass transaction volume. AP4M remains early infrastructure whose commercial impact will depend on real adoption and interoperability.

Stablecoins are positioned as a practical payment rail

Miebach’s stablecoin argument is narrower than the idea that crypto will replace national currencies. He has described dollar-denominated stablecoins as useful for suppliers in high-inflation economies and for cross-border corridors where moving between currencies can be slow or unpredictable. A dollar can be represented by a stablecoin on one side of a transaction and converted back into fiat on the other.

That makes stablecoins complementary to Mastercard’s broader payments strategy. They can provide another way to move and settle value while Mastercard supplies identity, controls, network connectivity, and acceptance. In this framing, stablecoins could expand access to dollar-based payments rather than displace the dollar itself.

Open Standard broadens the stablecoin bet

Mastercard is also participating in the Open Standard stablecoin initiative alongside other large payments, banking, technology, and crypto companies. Reuters reported that the consortium included more than 140 businesses and planned a U.S.-dollar-pegged stablecoin called Open USD, with the project expected to go live later in 2026.

The significance for Mastercard is strategic: the company is not limiting its future to one payment instrument. It is preparing for a market in which card credentials, bank accounts, and dollar-backed digital tokens may all be used by the same software agent, depending on the transaction’s cost, speed, geography, and rules.

Mastercard is funding the strategy from a strong core business

The AI and stablecoin initiatives are being developed alongside, not instead of, Mastercard’s existing payments business. In its second-quarter 2026 results, the company reported $9.3 billion in revenue and adjusted earnings of $5.04 per share; coverage of the results said both figures exceeded analyst expectations. Stronger spending, cross-border volume, and value-added services supported the quarter.

That context matters. Agentic commerce is being treated as a long-term growth opportunity while Mastercard’s established network continues to generate the resources for investment. The company does not yet need AP4M to replace a failing business model.

The main unresolved questions

The roadmap is technically coherent, but its success is not guaranteed. The difficult questions are commercial and institutional:

• Will consumers trust agents to spend without approving every purchase?
• Can permissions and spending limits be enforced consistently across platforms?
• How will merchants handle disputes when an agent misunderstands a user’s intent?
• Can different agent protocols and payment networks interoperate?
• What fraud, liability, and regulatory rules will apply when software—not a person—initiates the payment?
• Will the economics of sub-cent transactions justify the infrastructure required to process them?

Mastercard’s bet is that trusted identity and programmable authorization will become as important to agentic commerce as card acceptance is to human commerce. Agent Pay addresses the first problem—an AI assistant buying on behalf of a person. AP4M targets the next one: autonomous systems buying resources from one another continuously.

For now, the clearest conclusion is not that machines have replaced cards or dollars. It is that Mastercard is trying to make its network usable by machines while keeping verification, permissioning, tokenization, and settlement at the center of the transaction.