Stripe questioned it. The company processes online payments for millions of businesses and handles more than $1.9 trillion a year. Its question was a different one: how to approve more of the good customers the wall turns away without letting the bad ones through. That needs something a wall lacks: the ability to read a single payment and tell a first-time buyer from a thief using a stolen identity.

Radar, Stripe’s fraud prevention solution, has learned from the network for over a decade. Its advantage is what it has seen. When a card reaches a Stripe checkout, there is a 92% chance Radar has seen it before somewhere on the network, at another merchant or in another country. A single business sees only its own customers, so a card testing stolen numbers across a hundred sites looks new to it. Radar sees all hundred at once. It scores each payment against hundreds of signals in real time, which lets it block the riskiest and cut fraud by roughly a third on average. Stripe runs a second AI product beside it, Stripe Identity, which matches a customer’s government ID to a live selfie to stop fake accounts before they ever transact. But even with that reach, seeing is not understanding, and the newest payment attacks no longer match known patterns.