Alchemy Launches AgentCard: AI Agents Can Pay Directly Within Mastercard's Acceptance Network

Alchemy has launched a product called AgentCard, which, according to its announcement, allows AI agents to make payments directly at any merchant that accepts Mastercard online. This means agent payments are no longer confined to proprietary protocols or crypto-native settlement rails, but are now plugged into a traditional card network covering tens of millions of online merchants.

From Protocol to Card Network: The Shifting Path to Agent Payment Adoption

Over the past year, discussion around AI agent payments has largely centered on the protocol layer—x402 attempted to provide a standard interface for machine-to-machine payments using the HTTP 402 status code, while MCP addressed the context problem of agents invoking tools and data. But these protocols solve "how an agent initiates a payment instruction," not "which clearing channel the money ultimately travels through." AgentCard's entry point is precisely here: it maps payment requests initiated by agents onto Mastercard's online acceptance network.

According to Alchemy's announcement, AgentCard targets agent purchases at online merchants, rather than in-store card swipes or ATM withdrawals. This limitation is worth noting—online acceptance means the entire transaction is digital, requiring no POS terminal upgrades on the merchant side and no physical card on the agent side. For merchants, what they receive is still a standard card network transaction; for agent operators, the challenge is how to translate an agent's purchase intent into an authorization request acceptable to the card network.

Authorization and Risk Control: The Real Difficulty of Agent Payments

The core risk logic of card networks is built on "authorization by the cardholder personally." When the entity initiating payment shifts from a human to an AI agent, this logic faces a direct challenge: Who authorized this transaction? What is the scope of authorization? If an agent is manipulated through a prompt injection attack to overspend or pay a malicious recipient, how is liability determined?

AgentCard's effectiveness depends on how it handles these questions. From the perspective of Agent payment compliance and security that OceanAlt focuses on, the key is not whether an agent can initiate a payment, but whether there is an enforceable interception mechanism before settlement—including verification of agent identity (KYA, Know-Your-Agent), constraints on per-transaction and daily cumulative limits, whitelist management for recipients, and real-time identification of anomalous transaction patterns. If these capabilities remain only on the agent operator's side, with no corresponding verification on the card network side, the risk will surface at the clearing stage.

Mastercard has previously made moves in the agent payment space, including collaborating with multiple institutions to explore agent-initiated card transactions. AgentCard's integration into its acceptance network means agent payments are beginning to enter card organizations' existing risk control frameworks. However, card organizations' existing anti-money laundering (AML) and fraud detection models are primarily trained on the spending behavior of human cardholders. Agent transactions may differ significantly in frequency, amount distribution, and merchant selection patterns. Whether these models can effectively distinguish "normal agent procurement" from "agent abuse" currently lacks public data support.

Practical Implications for Stakeholders

For merchants, accepting agent payments requires no additional integration—as long as the transaction falls within Mastercard's online acceptance scope, it is treated as a standard card transaction. This lowers the barrier for merchants to participate in the agent economy, but also means merchants may receive orders initiated by agents without full awareness. Whether refunds, dispute resolution, and chargeback processes are suited to agent scenarios is an operational detail to watch going forward.

For agent developers and operators, AgentCard offers a more direct path than building proprietary stablecoin settlement rails: no need to handle on-/off-ramp, no need to hold crypto assets, and agents can complete purchases. The trade-off is accepting the card network's authorization rules and dispute resolution mechanisms, with agent operators bearing corresponding compliance responsibilities.

For card organizations and payment service providers (PSPs), the incremental transaction volume brought by agent payments is an opportunity, but only if risk control systems can cover new types of transacting entities. If agent transactions prove to have significantly higher fraud rates than human cardholder transactions, card organizations may tighten acceptance rules, which would instead slow the scaling of agent payments.

AgentCard's launch is a step in agent payments moving from protocol experiments to real card networks. It does not solve all problems—standardized expression of authorization intent (mandate), cross-platform agent identity verification, and a unified interface for pre-settlement interception remain infrastructure the industry needs to collectively build. But at least agent payments are beginning to enter a network with decades of accumulated risk control experience, rather than circulating within a closed sandbox. The direction of this step is more noteworthy than the number of scenarios it can currently cover.

---

Original source: Yahoo Finance · https://news.google.com/rss/articles/CBMinwFBVV95cUxPcDh4dnpzYmFSeXdMQkdvSlU1X1o2NkNuSVpyQk9UdUFXRVFSOVdFV0h6a2lQMnJ2cENLM25QbG5XY1VpSWhNekthcjR3Mk1QNUdaS29UdzdJbDhYSVRBemdFaFRUdVo0em1zQ0dPN0lWeG9xNWc1dy11NmFleDhTMUxaaE9TeFZFRkZCTm9oa0Y1eGVrTVRVaG1EREFCSXM?oc=5