At 2:14 a.m., a credit card that usually buys groceries in Ohio is used to purchase electronics in Romania. The bank's fraud detection system flags it before the transaction completes. A text message goes to the cardholder. The charge is declined. That sequence happens millions of times a day, and it relies on pattern recognition rather than human review.

Modern fraud detection combines rules and machine learning. Simple rules catch obvious problems: a transaction in a country you have never visited, three gas station charges in ten minutes, a purchase far above your usual spending. Machine learning models go further. They learn each customer's normal behavior and score every transaction for risk. If the score crosses a threshold, the bank acts.

The tools include real-time transaction monitoring, device fingerprinting, geolocation checks, and behavioral biometrics such as how you type your password. Banks share data through consortiums to spot fraud rings that move between institutions. None of it is perfect. False positives annoy customers and cost merchants sales. Fraudsters adapt constantly.

The stakes keep rising. Global card fraud losses exceed $30 billion a year, and scams involving authorized push payments are growing faster than card fraud. Banks now warn customers in the app before they send money to a suspicious account.