Mastercard Launches Payments for Machines, Built for Sub-Cent Transactions Between AI Agents
Agent Pay for Machines extends the card network to tiny, automated payments, with Stripe, Cloudflare, Coinbase and Checkout.com on board.
Key takeaways
- Mastercard's AP4M handles automated payments between software agents, some below one cent.
- Rules and spending limits are set up front and enforced programmatically, with agent credentialing for verification.
- Ant International, Checkout.com, Cloudflare, Coinbase and Stripe are named partners.
Mastercard has launched Agent Pay for Machines, or AP4M, a service for small automated payments between software agents and systems. Some of the transactions it targets are worth less than a cent, a size that card networks have historically treated as not worth processing.
What AP4M does
AP4M runs across Mastercard's global network and builds on its existing Agent Pay program, which launched in April 2025 for AI agents buying on behalf of people. The new service shifts the focus from shopping to machine-to-machine commerce: an agent paying for an API call, a data lookup, a slice of compute or a page of content, thousands of times a day, without a person approving each one.
Two features carry the design. The first is agent credentialing, which lets organizations verify that an agentic entity is who it claims to be. The second is programmable authorization: a company sets spending limits and rules in advance, and the system enforces them automatically as payments occur. Mastercard's Jorn Lambert described the target workload as "very high volumes, very small values, very fast and at extremely low latency."
Named partners are Ant International, Checkout.com, Cloudflare, Coinbase and Stripe. Cloudflare's presence is notable because it also worked with Visa on that company's Trusted Agent Protocol, which launched in October 2025 to help merchants tell legitimate commerce agents from malicious bots. Coinbase's involvement points to a link with stablecoin payments, which are already used for small transfers between programs. Mastercard has not published transaction volumes or revenue figures for AP4M.
Why sub-cent payments are hard
Cards were designed for human purchases. Fixed per-transaction costs make a fraction-of-a-cent charge uneconomic, which is why web publishers settled on subscriptions and ads rather than pay-per-page. If AI agents begin consuming content and services by the request, the old pricing models strain. AP4M is Mastercard's bet that its network can serve that traffic rather than cede it to stablecoins, direct blockchain payments or proprietary credits.
The programmable-limit approach also answers a risk specific to agents. A software agent that runs unattended can loop, be tricked by a malicious prompt, or simply overspend. Hard caps tied to a verified credential give a company a way to say what an agent can spend, with whom and for how long, before it ever runs. That is the same logic behind agentic tokens in the earlier Agent Pay product, which bind an agent to an individual user.
The competitive picture
Visa is pursuing the same market. Its Trusted Agent Protocol arrived in October 2025, and in December 2025 it said that hundreds of controlled agent transactions had run in live settings, with more than 20 agents or agent platforms integrating with Visa Intelligent Commerce. Visa predicted millions of consumers would use AI agents to complete purchases by the 2026 holiday season. Stripe, a partner of both networks, has also extended shared payment tokens to support them. The networks are racing to define identity and permission standards before agent traffic scales, because whoever sets those standards controls the trust layer.
Who the early users are
The partner list hints at the first use cases. Cloudflare sits in front of a large share of websites and can meter or block bots, so a payment layer lets sites charge agents that crawl or call their data rather than simply refusing them. Stripe and Checkout.com are payment processors that already serve developers and merchants, and can expose AP4M acceptance without merchants rebuilding their checkout. Coinbase brings wallet infrastructure and an existing push into small on-chain payments, while Ant International adds reach in Asia, where Mastercard says it ran its first live Agent Pay transaction in Hong Kong in late March, according to a secondary source.
None of the companies has said how many agents or merchants are live, and the Biometric Update report cites no volumes. That is typical of the early phase of a new rail, where announcements outrun traffic. The useful measure over the next year is not the number of partners but the number of transactions that settle without a human clicking approve.
What to watch
- Real volumes. Neither network has published agent transaction totals; the first disclosed figures will show whether this is a pilot or a product.
- Fees. Pricing for sub-cent payments will decide whether the card rail can compete with stablecoin micropayments.
- Holiday 2026. Visa's prediction of millions of agent shoppers gets its first test in the coming weeks.
- Liability rules. Who pays when an agent misbehaves remains unsettled across networks, merchants and consumers.
Sources
- Mastercard's AP4M enables small agentic payments at machine speed and scale — Biometric Update
- Visa Introduces Trusted Agent Protocol: An Ecosystem-Led Framework for AI Commerce — Business Wire
- Visa and Partners Complete Secure AI Transactions, Setting the Stage for Mainstream Adoption in 2026 — Nasdaq / Visa press release
