Mastercard rolls out AI payment option as card companies race for standards

Mastercard just gave your future AI assistant a wallet. The payments giant announced a partnership with Alchemy to launch a consumer-facing agentic payments option, letting cardholders hand a virtual Mastercard to an AI agent and authorize it to buy things online without requiring a human thumbs-up before every purchase.

Users set the guardrails. Spending caps, merchant restrictions, and optional pre-checkout approval are all baked in.

How the system actually works

The new option builds on Mastercard’s existing Agent Pay program, which the company originally introduced in April 2025 to bring tokenization into AI-driven commerce. At the core of the system are what Mastercard calls “agentic tokens,” dynamic credentials designed specifically for machine-initiated transactions.

Rather than storing a static card number, each AI agent receives a one-time-use or session-limited credential. That means if the token gets intercepted, it’s essentially worthless for any subsequent transaction.

Cardholders can configure their agents to operate within tight parameters. Want your AI to only buy groceries from two approved retailers and never exceed $200 per week? Done. Need it to book flights but require your sign-off on anything over $500? Also done.

Visa is running the same race

Mastercard isn’t operating in a vacuum. Visa has been building its own parallel infrastructure for AI-driven payments, having previously partnered with Alchemy and launched its Intelligent Commerce Connect platform.

The two payment networks have also been collaborating with Ant International on frameworks for verifying and monitoring AI agents across different networks.

The trust problem is real

For all the engineering sophistication, the biggest obstacle isn’t technical. Visa’s own research found that only 23% of US consumers trust generative AI to manage their payments. That’s less than one in four people willing to let a bot handle their money.

Both Mastercard and Visa executives have acknowledged this publicly, emphasizing that the success of agentic commerce depends on establishing what they call “verifiable intent,” essentially proving to the consumer that the AI did exactly what it was told to do and nothing more.

Risk frameworks that enforce user-defined parameters are supposed to be the answer. If the agent can only operate within boundaries the cardholder explicitly sets, the damage from any single error is contained.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.