Definition
Agent Pay for Machines (AP4M) is a Mastercard service that lets software agents pay one another across Mastercard’s network without a person clicking “buy.” Mastercard announced it on June 10, 2026. The premise is narrow but far-reaching: when an AI agent needs to purchase something — compute time, a data feed, a domain name, a single API call — AP4M handles the credentialing, permission checks, and settlement so the payment clears at machine speed, sometimes for amounts smaller than a cent.
It builds on Mastercard Agent Pay, which the company introduced in 2025. The two aren’t interchangeable. Agent Pay defines how a trusted agent acts on a person’s behalf in a fairly conventional purchase — picture an agent buying a sweater inside ChatGPT, with a human intent behind it. AP4M targets the layer underneath: the constant, automated back-and-forth between systems that nobody watches in real time. Mastercard’s chief product officer, Jorn Lambert, framed the goal as letting services be bought and sold among agents at scales today’s payments can’t reach — very high volume, very small value, very low latency.
Mastercard describes four capabilities holding it together. Every agent is credentialed and can be recognized across ecosystems using Verifiable Intent. Organizations set authorization rules and spending limits that get enforced programmatically. Verified participants connect and transact across providers. And settlement is guaranteed across multiple rails — cards, bank accounts, and stablecoins. The product page collapses that into a simpler loop: authorize, discover, execute, settle.
How it relates to marketing
For marketers, AP4M matters less as a payment rail and more as a signal about where buying decisions are moving. When agents transact on their own, the customer touchpoint moves with them. A brand’s homepage, its checkout flow, its retargeting pixel — all of it assumes a person is looking at a screen. An agent buying cold-chain monitoring data for a shipment isn’t looking at anything. It reads structured information and acts within a budget.
That changes two things at once. First, the unit of commerce gets smaller. If an agent pays a tenth of a cent for one data lookup and does that ten thousand times an hour, the old model of “drive a visit, convert a cart” stops describing the transaction at all. Marketers who sell data, content, compute, or APIs may find their product becoming something agents meter rather than something humans subscribe to.
Second, discoverability shifts from the human web to the agent web. An agent picking a “trusted service provider” — Mastercard’s own coffee-shop example has an agent choosing imagery, hosting, and a domain registrar — is making a vendor selection your brand either qualifies for or doesn’t. Being legible to agents, credentialed, and priced for machine consumption becomes its own kind of positioning. This is the same concern that sits behind Share of Model and Agent Readiness, applied to the moment money actually changes hands.
How It Works
There’s no formula to calculate here; AP4M is infrastructure, not a metric. What’s worth understanding is the sequence, because the mechanics explain why marketers and merchants care.
Take Mastercard’s flower-shop scenario. An entrepreneur tells an agent to build and launch the shop’s web presence inside a set budget. The agent finds providers for imagery, content, a domain, and hosting, then puts together a plan with an estimated cost. The business approves it by setting a spend limit against a funding source — a card, a stablecoin wallet, or a credit line — and that authorization gets recorded in an on-chain smart contract. The agent then executes, paying each provider with verifiable vouchers, which are off-chain credentials that move fast and cheap. Providers batch those vouchers and submit them for settlement, taking payout in whatever currency they prefer, fiat or stablecoin.
The split between on-chain authorization and off-chain execution is the clever part. Recording every fraction-of-a-cent payment on a blockchain would be slow and expensive. So the spending authority lives on-chain, where it’s auditable and binding, while the individual micropayments happen off-chain through cryptographic verification and get aggregated for settlement later. Mastercard’s network sits over the top, supplying the credentialing, the controls, and a settlement guarantee so a provider knows it’ll actually get paid.
Verifiable Intent is the trust anchor. It’s the artifact proving an agent acted inside the authority a human or business granted it. Without something like that, a merchant has no way to tell an authorized agent from a runaway script, and no basis for resolving a dispute later.
How to Utilize
The use cases Mastercard and its partners point to fall into a few buckets.
Agent-built services. A business hands an agent a goal and a budget, and the agent assembles the pieces by buying from multiple vendors. The flower shop and coffee shop examples both work this way: one human request turns into a chain of purchases across providers, executed automatically.
Machine-to-machine operations. A logistics agent managing a delivery route pays for freight, reserves a loading bay, buys temporary cold-chain monitoring, and settles warehouse handling fees as a shipment moves. No single payment is large. The volume and the timing are what make it hard for human-era rails.
Metered consumption of compute and data. This is the one several partners emphasize. Agents pay per API call, per browser session, per inference. Nevermined’s co-founder put it plainly — metered pricing is the business model of AI agents, and metering only works if the agent can pay and get paid. Cloudflare’s pitch is similar: agents built on its platform now need a trusted way to pay for the resources they burn through.
For a marketer or product owner, the practical question is whether your offering can be sold this way. If you publish data, run an API, license images, or sell any digital service in small increments, AP4M is a route to selling it to agents at a price point that wouldn’t survive a traditional checkout. If your product is a considered human purchase — a car, a mortgage, a B2B contract — this isn’t your rail, though the consumer-facing Agent Pay program might be.
Comparison to Similar Approaches
AP4M is one entry in a crowded field of agentic and machine-payment efforts that took shape between 2024 and 2026. They overlap, and several are designed to interoperate rather than compete outright.
| Approach | What it is | Who initiates | Settlement | Trust / identity layer |
|---|---|---|---|---|
| Mastercard AP4M | Network service for machine-to-machine and micropayments | AI agents, autonomously | Multi-rail: cards, accounts, stablecoins | Network credentialing + Verifiable Intent |
| Mastercard Agent Pay | Framework for trusted agents transacting on a person’s behalf | Agent acting on human intent | Mastercard rails | Agentic Tokens + Verifiable Intent |
| Visa Intelligent Commerce / TAP | Visa’s parallel framework; agents present credentials to merchants | Agent on consumer’s behalf | Visa rails | Trusted Agent Protocol credentials |
| Stripe Machine Payments Protocol (MPP) | Open standard, co-authored with Tempo, for agents to pay over HTTP | Agents | Stablecoins, cards, BNPL via Shared Payment Tokens | Shared Payment Tokens |
| Google AP2 | Vendor-neutral mandate format, donated to the FIDO Alliance in 2026 | Agent, under signed mandate | Rail-agnostic | Cryptographically signed Mandates |
| x402 | Open protocol using the HTTP 402 status code for pay-per-request | Agents / clients | Stablecoins (originally) | Payment challenge + credential |
The lines blur on purpose. As of April 2026, payment providers can emit AP2 Mandates that Mastercard accepts as Verifiable Intent, and Stripe and Visa have both signed onto AP2 too. Coinbase, an AP4M partner, has tied its support to the x402 standard. So the practical reality isn’t one network winning — it’s a set of protocols that increasingly recognize each other’s credentials. AP4M’s distinguishing pitch is the settlement guarantee and the reach of an existing global card network behind transactions that are otherwise pure software.
Best Practices
If you’re a business or marketer evaluating AP4M, a few things are worth getting right early.
Set spending limits before anything else. The whole model depends on bounded authority — an agent should only ever be able to do what it was authorized to do. Treat the spend limit and the funding source as the first design decision, not an afterthought.
Make your service legible to agents. An agent can only buy from you if it can find you, verify you, and understand your pricing without a human reading a marketing page. Structured pricing and a clean machine-readable interface do more here than persuasive copy.
Price for metering, not just for subscriptions. Selling in fractions of a cent is a different exercise than selling a monthly plan. Work out whether your margins survive per-call pricing before you expose a service to high-frequency agent demand.
Keep an audit trail. Disputes are still unsettled territory in agent commerce — t54 Labs and others are building transaction-level risk and traceability precisely because chargebacks and liability get murky when no human pushed the button. Capture authorization records and receipts so you can answer “who approved this” after the fact.
Don’t bet on a single protocol. Given how much the major efforts are converging on shared mandate formats, designing your payment edge so it can accept AP4M, AP2, MPP, or x402 credentials protects you from picking wrong.
Future Trends
Interoperability looks like the dominant direction. The donation of AP2 to the FIDO Alliance and Mastercard’s acceptance of AP2 Mandates as Verifiable Intent both point toward a world where one agent credential works across networks, rather than a walled-garden fight. That’s unusual this early in a category, and it’s probably driven by everyone wanting the agent economy to actually function.
Stablecoins are moving from edge case to default option. AP4M settles across cards, accounts, and stablecoins on equal footing, and a striking share of its launch partners — Coinbase, Circle-adjacent infrastructure, BVNK, Ripple, Solana Foundation — are there specifically for the stablecoin rail. RippleX’s executive called regulated on-chain stablecoin settlement an emerging enterprise standard. Whether that holds depends on regulation that’s still being written.
Liability and disputes are the unresolved problem. When an agent orders the wrong thing or overspends, the four-party dispute model that card networks have used for decades gains a new participant with no clear legal standing. Expect a lot of activity around “Know Your Agent” verification, evidence layers, and dispute frameworks before machine payments scale to the volumes Mastercard imagines.
And the addressable surface keeps widening. Mastercard enabled its consumer Agent Pay across all US cardholders by late 2025 and has been adding markets like Hong Kong through 2026. AP4M extends that reach into the machine layer. The plausible end state is most transactions never touching a person at all — Alchemy’s co-founder described an economy where machines pay each other constantly for things too small to bother a human with.
FAQs
Is AP4M the same as Mastercard Agent Pay? No. Agent Pay covers a trusted agent making a purchase on a person’s behalf, often a normal-sized consumer transaction. AP4M is built for the automated, high-frequency, often tiny payments that happen between systems in the background. AP4M builds on the Agent Pay program rather than replacing it.
Can an agent spend without limits? No. Authorization rules and spending limits are set upfront and enforced programmatically. In Mastercard’s examples, a business approves a budget tied to a funding source, and that authority is recorded in an on-chain smart contract before the agent transacts.
What can be paid with — cards or crypto? Both. AP4M supports multi-rail settlement across cards, bank accounts, and stablecoins, and a provider can take payout in its preferred currency, fiat or stablecoin.
How small can a transaction be? Mastercard cites payments as small as fractions of a cent. The off-chain voucher mechanism exists precisely so micropayments at that scale clear quickly without per-transaction blockchain costs.
Who’s actually using it? At launch Mastercard named more than 30 participants and supporters, among them Adyen, Ant International, Checkout.com, Cloudflare, Coinbase, Global Payments, Stripe, and Tempo. Many are validating specific use cases rather than running it at scale yet.
Does this compete with Stripe’s MPP or Google’s AP2? Less than you’d think. The standards are converging — Mastercard accepts AP2 Mandates as Verifiable Intent, and partners like Coinbase tie support to x402. AP4M’s differentiator is the settlement guarantee and the existing global network behind it.
What’s Verifiable Intent? It’s the credential that proves an agent acted inside the authority a person or business granted it. It’s how a counterparty trusts that an agent is permitted to make a given payment, and how authorization can be checked after the fact.
What happens when an agent makes a mistake? That’s genuinely unsettled. Dispute and liability rules for agent-initiated payments are still being worked out, which is why several partners are building Know Your Agent verification and transaction-level audit trails. Treat record-keeping as essential until the frameworks mature.
Related Terms
- Mastercard Agent Pay
- Agentic Commerce
- Brand Visibility for Agentic Commerce (BVAC)
- Verifiable Intent
- x402 Payment Protocol
- Machine Payments Protocol (MPP)
- Agent Payments Protocol (AP2)
- Shared Payment Tokens (SPTs)
- Visa Intelligent Commerce
- Multi-Agent System (MAS)
- Stablecoin
Sources
- Mastercard. “Mastercard launches Agent Pay for Machines to unlock super-fast, always-on payments.” https://www.mastercard.com/us/en/news-and-trends/press/2026/june/mastercard-launches-agent-pay-for-machines.html
- Mastercard. “Mastercard Agent Pay for Machines” (product page). https://www.mastercard.com/us/en/business/artificial-intelligence/mastercard-agent-pay/agent-pay-for-machines.html
- Mastercard. “Mastercard Agent Pay.” https://www.mastercard.com/us/en/business/artificial-intelligence/mastercard-agent-pay.html
- RisingWave. “Mastercard Agent Pay vs Visa Intelligent Commerce vs Stripe Agentic Commerce: 2026 Comparison.” https://risingwave.com/blog/mastercard-agent-pay-vs-visa-vs-stripe-agentic-commerce/
- Stripe. “Introducing the Machine Payments Protocol.” https://stripe.com/blog/machine-payments-protocol
- Google Cloud. “Powering AI commerce with the new Agent Payments Protocol (AP2).” https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-payments-protocol-ap2
- Crossmint. “Agent card payments compared: Visa, Mastercard, Stripe, Ramp and Slash.” https://www.crossmint.com/learn/agent-card-payments-compared
- American Banker / PaymentsSource. “Visa, Mastercard expand agentic AI deployments.” https://www.americanbanker.com/payments/news/visa-mastercard-expand-agentic-ai-deployments
- Cloudflare. “Agentic Payments.” https://developers.cloudflare.com/agents/agentic-payments
