Satya Vandrangi, an AI leader at AWS, is passionate about trust and safety in payments. NYC cyclist and food explorer.
Every financial system is a nightclub with two bouncers.
The first stands at the door. He checks your ID and decides whether you belong inside. In banking, we call this onboarding, KYC and account screening. It is why some people never get in at all.
The second works the floor, watching for the person already inside who is about to cause trouble. In banking, that is transaction monitoring, and it runs on every card swipe, every transfer and every deposit. Once you are inside a financial system, the greatest threat to your money is usually not the institution, but other patrons.
While almost everyone agrees with the purpose of the second bouncer, the question remains: What should the second bouncer cost, and who should pay?
The One Place The Price Is Visible
You can see the answer at a gas station. Cash, $3.29. Credit, $3.39. It is one of the last places where the cost of moving money is printed on the street where anyone can read it. Everywhere else it is folded into the shelf price, which means all of us pay it, all the time, without ever seeing the number.
That dime is not a markup. Part of it is rewards. Part of it is the guarantee that if something goes wrong, someone will make you whole. And part of it pays the bouncer on the floor.
Between your card touching the terminal and the word APPROVED, a system answers a question with real money riding on it. Is this person who they claim to be? Visa says its models examine more than 500 attributes of a transaction and return a risk score in milliseconds. Visa processed 258 billion transactions last year, over 700 million a day.
This is one of the most sophisticated machine learning systems on earth, and almost nobody thinks about it, because the entire product experience is a beep. Visa reports investing $13 billion in technology over five years and monitoring 22 billion security events a day. It has been at this since 1993, when it became the first payment network to deploy AI for risk management. Thirty years of investment in one capability: deciding, faster than a person can perceive, whether to trust a stranger.
A Technology Problem That Never Got Cheaper
Which tells you something important. It does not cost money to move money because moving money is hard. Moving money is nearly free. FedNow and RTP already do it for a few cents. It costs money because trusting the movement is hard.
Trusting the movement is a technology problem, and technology problems have a well-known property: They get cheaper. Every input to this system has fallen by orders of magnitude since 1993.
So why hasn’t the price?
There are legitimate reasons.
Fraud is the only technology problem with an opponent, and he rides the same cost curve we do, now with generative tools of his own.
But the deeper obstacle is a mental model. We still treat protecting money as a premium feature available on certain networks and not others, when it should be treated the way we treat the internet. Basic infrastructure, something everything else assumes is there and arguably closer to a right than a product.
Where There Is No Bouncer At All
Now consider the largest payment most people will ever make. You are buying a house, and you wire the down payment. That transaction carries no confirmation of who is on the other end, no dispute process and no reversal. It is final the second it lands.
In 2025, real estate wire fraud accounted for more than $275 million in reported losses and business email compromise for over $3 billion, most of it moving by wire or ACH, which is precisely why so little comes back.
Compare that to buying a cup of coffee. Roughly two percent of that four dollars buys fraud models, monitoring and the right to dispute the charge. On a $300,000 wire, you pay thirty dollars and get none of it.
We have built our most sophisticated protection around our smallest transactions and left our largest ones bare. It is not that nobody would pay for it. There is simply nothing to buy.
The Layer We Never Built
The internet’s great trick was to stop caring what you were doing. A packet is a packet, moving through the same rails at the same cost whether it carries a video call or a photograph of someone’s lunch.
It also solved the question we are still struggling with. The Transmission Control Protocol (TCP) guarantees your data arrives and resends what gets lost. User Datagram Protocol (UDP) throws packets and hopes. Both run on the same rails, and the application picks based on what the traffic is for. Nobody built a separate internet for reliable data.
Payments never made that leap. We still build a vertical stack for every kind of payment, each with its own rules and its own idea of what happens when things go wrong. Tap a card, and you get one set of protections. Send from your account, and you get another, usually none. The safety you receive has little to do with what you were buying or who you are. It depends on which pipe your money took.
Whether that means new rails or better use of the ones we have, I do not know. What I do know is that protection has to become something common that every rail is built on, not something you get or do not get depending on which one you used.
Trust is not a feeling. It is a prediction. People trust systems whose behavior they can anticipate, which is why cash still wins with millions of households. Cash has no pending state, no posting order, no hold and no bouncer at the door.
The work in front of us is not convincing people to trust institutions, but building institutions predictable enough to deserve it. Payments has been waiting a long time for its internet moment. Whether we like it or not, it still has not arrived.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?