Mastercard is opening early access to Agent Pay for Machines, a payment framework designed to let AI agents purchase digital services within approved budgets and permissions.

AI agents are moving beyond answering questions and generating content. They are beginning to carry out tasks that require access to paid services. Mastercard is now developing payment infrastructure to support that shift.

At Cloudflare Connect 2026, Mastercard demonstrated Agent Pay for Machines, a system designed to let AI agents pay for digital services on behalf of businesses while keeping spending authority under human control.

The initiative targets a growing challenge in the emerging agentic economy: how can an AI agent purchase data, computing power, AI model access, or other services without needing a separate account and billing arrangement for every provider?

Mastercard’s approach combines the open x402 payment protocol with its payment infrastructure and a system of digitally signed payment vouchers. The goal is to let agents request services, authorize payments, and continue working without receiving unlimited spending authority.

The company also announced early access for developers, providing APIs, tools, a sandbox, and documentation to help them build and test agent-to-agent commerce applications.

How Agent Pay for Machines Works

The system is designed around a simple principle: an AI agent can transact within a defined budget, but it does not receive unrestricted authority to spend.

A business first assigns an agent a task and an approved spending limit. When the agent encounters a paid digital service, the provider can respond with an HTTP “402 Payment Required” message.

The agent then creates a digitally signed payment voucher. The provider verifies the voucher before delivering the service. The agent can continue its task, while providers can combine vouchers and redeem them later.

According to Mastercard, this process is designed to support machine-speed interactions while separating the agent’s ability to perform a task from unlimited access to funds.

That distinction matters for businesses. An autonomous agent may need to select different services as a task unfolds. However, the business still needs controls over what the agent can buy, how much it can spend, and which services it can access.

A Real-World Demonstration With Hype

Mastercard demonstrated the system through a pilot involving Hype, a marketing agency that already uses AI agents in campaign work.

The demonstration gave an AI agent a task: plan a sample marketing campaign for a protein bar launch. The agent identified two services it needed to complete the work.

  • AIsa: for social media research.
  • Cloudflare AI Gateway: for generating campaign images.

Hype approved the funding within the defined task and budget. The agent then obtained the research, requested eight campaign images, and made payments as the services were used.

Mastercard described the demonstration as involving nine payments across two independent merchants, all within one approved budget.

The companies say the workflow did not require separate accounts, API keys, or invoices for each service. Instead, the agent used payment vouchers to access the services and continue its work.

The demonstration illustrates how agentic commerce could operate in practice. An AI agent could choose a service, pay for it, and use the result as part of a wider task.

However, this was a pilot demonstration. It does not establish that all AI agents or merchants can already use the system at scale.

The Technology Behind the Payments

The pilot combines several components from Mastercard and its partners.

The x402 protocol provides a way for online services to request payment when a user or agent tries to access paid content or functionality. Mastercard’s system builds on that mechanism to support a controlled payment workflow for AI agents.

Coinbase Developer Platform provides wallet infrastructure used to fund the agent, while Base supplies the on-chain settlement network. Mastercard says BVNK acts as the agent wallet, with Mastercard Global Money Platform supporting payment and settlement on both sides.

The architecture is intended to accommodate different payment preferences, including fiat currencies and stablecoins, depending on the relevant service and payment arrangements.

This approach reflects a broader industry effort to make machine-to-machine transactions more practical. AI agents may need to purchase small amounts of data, use a model for a specific task, or obtain computing resources for a limited period. Traditional account-based billing can add friction to these interactions.

Automated payment mechanisms could help simplify the process. Still, security, authorization, settlement, refunds, and dispute handling remain important considerations as these systems develop.

Mastercard Opens Early Access for Developers

Mastercard is making the first version of its buyer-side and seller-side APIs available through Agent Pay for Machines Early Access.

The program includes a toolkit, sandbox, and developer documentation. These resources are intended to help developers experiment with payment-enabled AI agents and services.

The company says it has conducted end-to-end pilots with Cloudflare, Ant International, AIsa, Base, BVNK, Coinbase, and Hype.

Early access gives developers an opportunity to test the framework and explore potential applications. The program also gives Mastercard and its partners a way to learn from real-world implementation needs.

For fintech companies, payment providers and developers building AI agents, the initiative raises a practical question: how should payment infrastructure adapt when software itself begins choosing and purchasing services?

Why Agentic Payments Matter for Fintech

The significance extends beyond a new payment API. AI agents may increasingly need to transact across multiple providers without direct human involvement at every step.

For example, a business agent might purchase market data, run an analysis through an AI model, and pay for additional computing resources. Each service could come from a different provider, with costs incurred as the task progresses.

A payment framework designed for agents could help connect these transactions. It could also let businesses set clear spending limits and permissions before an agent begins work.

Yet adoption will depend on more than payment speed. Businesses will need reliable identity and authorization controls, transparent records, clear liability arrangements, and safeguards against unauthorized spending.

Interoperability will matter as well. An agent economy cannot depend on every provider requiring a unique account, wallet, or payment integration. Shared protocols and compatible infrastructure could reduce those barriers.

Mastercard’s early-access program is an attempt to test this model with developers and ecosystem partners. The next challenge is proving that it works reliably across different services, platforms, and payment arrangements.

What Comes Next?

Agent Pay for Machines remains an emerging infrastructure initiative. Mastercard has invited developers to test the APIs and provide feedback as the roadmap develops.

Important questions remain around commercial availability, broader merchant adoption, transaction controls, and how disputes or failed services will be handled. The company has not provided evidence that the model has reached widespread production deployment.

Even so, the direction is clear. As AI agents take on more operational tasks, they will need ways to access paid services without creating excessive friction or losing human oversight.

Mastercard’s message is that the agentic economy needs more than capable AI systems. It also needs a trusted way for those systems to pay.

For fintech, the next stage of AI adoption may depend not only on what agents can do but also on how safely they can transact.

Source: What happens when AI agents start paying each other? Watch Mastercard Agent Pay for Machines in action