Every API call to the bank, inspected in under a millisecond.
This is a mock retail bank running behind Cloudflare. Below, watch five real attacks β
account takeover, data theft, fraud β succeed against an unprotected API, then get stopped
cold at the edge. Nothing is simulated: the scores and blocks are Cloudflare's.
Before a single line of bank code runs, the edge has already fingerprinted this request.
A bank uses exactly these signals to tell a customer from a bot.
Bot Management score
1 = definitely bot 99 = definitely human
Location
β¦
IP address
β¦
Network (ASN)
β¦
TLS
β¦
JA4 fingerprint
β¦
Verified bot
β¦
Threat score
β¦
Data centre
β¦
Five attacks on the bank
click Run to fire a real request
Each maps to a named banking risk. In monitor mode the attack
succeeds; in enforce mode the edge blocks it. Same traffic, one switch.
The fraud engine
custom Cloudflare Worker
Native features stop the commodity attacks with zero code. For bank-specific decisions,
one small Worker fuses three signals into a real-time allow / step-up / block. Move the
dials β the decision is computed live by the deployed Worker.
90
human β β bot
geo
account home is US β anything else is a geo mismatch
1
per 30s window
Fraud risk
ALLOW
Why this matters to the bank
π
Fraud & ATO
Automated credential stuffing and takeover stopped before login logic runs β no fraud-ops ticket, no customer churn.
π
Compliance & privacy
Broken object-level authorization is the #1 API risk. Blocking cross-customer reads at the edge is a reportable-breach avoided.
π
Know your attack surface
API Shield discovers undocumented endpoints automatically β you can't audit what you can't see.