This was the week agentic payments stopped being theoretical. Santander and Visa ran live transactions initiated by AI agents across Latin America. Coinbase turned a 29-year-old HTTP status code into a stablecoin settlement protocol, and Cloudflare, Google, and Visa joined the foundation behind it. FedEx committed to embedding AI agents in over half its operations by 2028.
At the same time, the risks sharpened. Nasdaq's Verafin unit reported global fraud losses climbed 9.2 percent in 2025, driven by the same AI capabilities powering legitimate commerce. A federal judge drew the first legal line on AI shopping agents. And Anthropic fought the Pentagon, reset its pricing, and shipped visual intelligence, all in a single week.
The throughline is not subtle. The payment layer, the settlement layer, the trust layer, they are all being rewritten simultaneously. And the companies doing the rewriting are not waiting for permission.
The question is no longer whether AI agents will move money. It is who controls the rails they move it through.
Payments
Santander and Visa Run the First Live Agentic Payments
Santander completed live agentic payment transactions with Visa across multiple Latin American markets. Real money, real rails, transactions initiated by AI agents on production networks. Days earlier, JP Morgan announced similar capabilities with Mirakl. The sandbox era for agentic payments is over. The question now shifts to scale, liability, and who owns the identity layer when the buyer is software.
AI Is Driving a Global Fraud Surge. The Numbers Are Getting Worse.
Global fraud losses climbed 9.2 percent in 2025, according to Nasdaq's Verafin unit. Generative AI has made phishing, deepfake identification documents, synthetic voice fraud, and automated social engineering scalable. Bold Security launched with $40 million for AI endpoint security. Sardine and Sift are deploying machine learning across transaction networks. But every new AI-powered payment channel is also a new AI-powered attack surface.
x402: The Internet Just Got a Payment Button. AI Agents Are Already Using It.
HTTP 402 "Payment Required" existed since 1997 with no implementation. Coinbase built x402 to turn it into a stablecoin settlement protocol for AI agents. The x402 Foundation, co-founded with Cloudflare, now includes Google and Visa. Over 15 million transactions processed, but CoinDesk reports real daily volume is roughly $28,000. The infrastructure is ahead of the demand, which is either a bet on the future or a warning sign.
AI
Anthropic's Week: A Pentagon Blacklist, a Pricing Reset, and Visual Intelligence
Anthropic was blacklisted by the Pentagon for building AI that was deemed "too ethical," sued the US government in response, eliminated long-context pricing surcharges making requests over 200,000 tokens dramatically cheaper, and shipped inline visual intelligence for Claude. Microsoft filed an amicus brief in support. Employees from OpenAI, Google, and DeepMind signed a public letter backing Anthropic's position. One company's week reshaped the debate on AI safety, pricing, and government procurement.
MCP vs Agent Skills: The Two Layers Every AI Team Is Confusing
The Model Context Protocol (MCP) provides infrastructure access, connecting agents to external systems via a standardised interface. Skills provide behavioural guidance, teaching agents what to do with that access. Most teams building agents are investing heavily in one layer and missing the other entirely. The result is agents with powerful connections but poor judgment, or sophisticated reasoning with no way to act on it.
Commerce
FedEx Is Building an AI Agent Workforce. Half Its Operations Will Run on It by 2028.
FedEx described the most specific enterprise agentic deployment plan we have seen. A three-tier hierarchy: manager agents that oversee workflows, worker agents that execute tasks, and audit agents that verify outcomes. Over 50 percent of operational workflows by 2028, across 220+ countries. FedEx touches nearly every e-commerce transaction. When their logistics stack goes agent-powered, every retailer and marketplace that ships through them interacts with AI agents whether they chose to or not.
Amazon Just Drew the Legal Line on AI Shopping Agents
A federal judge blocked Perplexity's Comet agent from placing orders on Amazon without the platform's authorisation. This is the first time a US court has formally intervened in the question of whether AI agents can shop on platforms without consent. The ruling establishes that platforms control which agents can transact on their systems. Every company building autonomous shopping tools needs to read the implications carefully.
That is seven stories across payments, AI, and commerce. The connecting thread: the infrastructure that powers how money moves online is being rebuilt for a world where the buyer, the seller, and the intermediary might all be software. The companies writing the protocols today are writing the rules for a decade.
If you found this useful, share it with someone who should be reading it.
— Major Matters
Which of these stories will still matter six months from now? Hit reply and tell us.
Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.