We found card testing on a checkout by accident. A merchant's approval rate fell from 81% to 64% in two days. The first guess was an issuer problem or a routing bug. It was neither.
The logs showed about 3,000 attempts an hour from a few hundred IPs. Each attempt was a $1 authorization with a different PAN, and almost all were declined. A few percent were approved, and that was the point: the attacker was checking which stolen cards were still live.
The checkout endpoint had no limit on failed attempts per session, per IP, or per BIN range. It accepted anonymous POSTs and returned a clean decline code each time. We had effectively built a card validity oracle and handed it out for free.
The fix was boring:
- Cap failed attempts per session and per device fingerprint, not just per IP. Botnets rotate IPs.
- Add a CAPTCHA or 3DS step after the second decline.
- Return the same generic message for every decline reason on the client side. Keep the real code in your logs.
- Alert on decline ratio per hour, not per day. A daily average hides a four-hour burst.
The scheme-side cost matters too. Acquirers and card networks watch your decline ratio and can flag you for it, even if every card was stolen by someone else.
Where do you put the first rate limit: at the edge, in the app, or at the gateway? Which layer has worked for you?
Payneteasy handles this kind of payment flow for merchants and PSPs — see how it works.



