For all the talk of AI agents that can buy things, the payments industry has been cautious. Pilots ran in sandboxes. Announcements came with caveats. The phrase "proof of concept" appeared in every press release.
This week, Banco Santander moved past all of that. The bank conducted its first live agentic payment transactions with Visa across multiple Latin American markets. Real money. Real networks. Transactions initiated by AI agents, not humans clicking buttons.
Agentic payments just crossed from concept to production. The question now is how fast the rest of the industry follows.
What Santander Actually Did
The details matter. Santander did not run a single test transaction in a controlled environment. It conducted live transactions across multiple markets in Latin America, using Visa's network infrastructure.
This means the full payment stack was engaged: authorisation, clearing, settlement. An AI agent initiated a payment request, that request flowed through Visa's rails, and a real merchant received real funds. The operational complexity of making that work, across multiple jurisdictions with different regulatory frameworks, is significant.
Latin America is a deliberate choice of testing ground. The region has a fragmented payments landscape with high mobile adoption and fast-moving fintech regulation. If agentic payments can work across multiple LatAm markets, the operational model can likely scale elsewhere.
Santander was not alone this week. Days earlier, JP Morgan Payments announced a partnership with Mirakl, the marketplace commerce platform, to build a joint solution enabling autonomous payments by AI agents.
Mirakl powers marketplace infrastructure for enterprise retailers. JP Morgan Payments processes trillions in annual transaction volume. The two companies building an agent-ready payment solution together signals that this is not a niche experiment. The largest payment processors in the world are engineering their systems for AI-initiated transactions.
Two of the world's most significant financial institutions, within days of each other, making concrete moves on agentic payments. That is a pattern, not a coincidence.
The Card Network Race
The card networks have been positioning for this moment. Visa is now running live agentic transactions in emerging markets. Mastercard has been developing its Verifiable Intent framework, a system for cryptographic proof that an AI agent has genuine authorisation to spend on a consumer's behalf.
The approaches are different. Visa is moving fast with live deployment, proving the operational model works. Mastercard is building the trust layer, focusing on the authorisation and verification challenges that get harder at scale.
Both approaches are necessary. Live transactions prove the concept. Trust frameworks make it sustainable. The network that gets both right first will likely become the default rail for agentic commerce.
The Authentication Problem Gets Real
As we examined in Who Authorised the Agent?, the core challenge of agentic payments is not technical. The card networks can process the transactions. The challenge is legal and structural.
When a human taps a card, the liability chain is clear. The cardholder authorised the transaction. But when an AI agent initiates a payment on a consumer's behalf, new questions emerge:
Did the consumer genuinely authorise this specific transaction, or just a general class of transactions?
If the agent was manipulated through prompt injection into making an unintended purchase, who bears the loss?
How does a merchant verify that an AI agent has the authority it claims?
These questions were theoretical when agentic payments were in sandboxes. With live transactions flowing through Visa's network, they require answers. Dispute resolution, chargeback frameworks, and liability allocation all need updating for a world where the entity initiating the payment is not the entity bearing the financial consequence.
Why Latin America First
The choice of Latin America is strategic beyond regulatory pragmatism.
The region is one of the world's fastest-growing digital payments markets. According to the FSB's cross-border payments reform agenda, international payment friction remains a systemic issue, and LatAm's fragmented landscape makes it an ideal proving ground for new payment models.
Santander has deep regional presence across Brazil, Mexico, Chile, Argentina, and Uruguay. Testing agentic payments across these markets simultaneously gives Visa and Santander data on how agent-initiated transactions behave across different currencies, regulatory regimes, and consumer banking systems.
If the model works here, it works most places.
What to Watch
Transaction volume data. Live transactions are a milestone, but scale matters. Watch for disclosures on how many agentic transactions are flowing and what the approval and dispute rates look like compared to human-initiated payments.
Network rules updates. Visa and Mastercard will need to update their network rules to formally accommodate AI-initiated transactions. How they classify these transactions, and what liability frameworks they attach, will shape the economics for everyone in the chain.
Regulatory response. Central banks and payment regulators in Latin America will be watching closely. If agentic transaction volumes grow, expect new guidance on consumer protection, agent authorisation requirements, and cross-border treatment.
Sources
When AI agents can initiate payments across borders and currencies, are we ready for the disputes that will follow?