Cross River Bank and Stripe formally partnered to introduce bank-grade card issuance infrastructure for autonomous AI agents. The collaboration enables virtual, single-use cards, allowing software to spend money without accessing a user's underlying payment credentials.
Autonomous AI agents—programs designed to plan, act, and complete tasks without human approval at each step—have operated in production for two years. Until recently, these agents could not autonomously pay for services. Stripe's architectural thesis, unveiled at Stripe Sessions 2026 on April 29, states that human-centric payment infrastructure cannot serve machines.
Modern payment systems assume a human is always involved. Credit cards require a person to enter a billing address, and subscription models align with human decision rhythms via monthly cycles. Even advanced payment APIs presuppose human initiation or approval for money transfers.
AI agents expose the limitations of these assumptions. An agent consuming API tokens at millisecond intervals cannot wait for a month-end billing statement. A multi-agent workflow, where one AI hires another for a subtask, cannot use a credit card for payment.
The economic model for AI products presents further challenges. Businesses incur compute costs instantly as tokens are consumed, but often collect revenue on a monthly lag. The ideal payment model requires real-time, per-token settlement.
Existing card rails are insufficient for this granular, high-velocity demand. Transaction fees alone would exceed payment values at sub-cent scales. The sheer volume of millions of settlements per second across a large AI product would overwhelm any billing system built for human purchasing patterns.
Stripe addresses these issues with its Agentic Commerce Suite and Agentic Commerce Protocol. The suite helps AI agents locate Stripe merchants, while the protocol establishes an open standard for agents and merchants to exchange information regarding availability, pricing, and fulfillment. OpenAI co-developed the protocol.
This integration allows agents to complete checkout using a shared technical language. Even with agent-initiated purchases, the seller remains the merchant of record. Stripe processes these payments through its existing compliance and dispute resolution frameworks, allowing agents to transact with established businesses without needing separate marketplaces or trust systems.
Stripe holds an advantage due to its established position processing payments for millions of businesses. Machine-to-machine transactions appear in the same dashboards merchants already use, resulting in near-zero switching costs for existing customers. Visa is also developing its own strategy, extending machine-to-machine payments to card-based systems, allowing agents to use existing financial infrastructure without requiring merchants to adopt crypto or stablecoins.
As enterprises use multiple AI models, payment rails, and liquidity sources, software determines the optimal supplier for demand based on cost, performance, risk, and availability. The most valuable infrastructure layer may coordinate these choices rather than owning every underlying product.
While Stripe leverages traditional finance rails for this infrastructure, the core demand for real-time, high-frequency, sub-cent settlement aligns with capabilities native to on-chain payment protocols. Stablecoins and DeFi lending platforms offer the programmable, continuous settlement mechanisms that could facilitate machine-to-machine commerce, presenting an alternative to card-based systems for agentic payments.
