A developer says “I want to build a site like DoorDash — let me copy this example.” Two risks hide in that sentence. Here are the six Cloudflare controls that cover both.
“Is this reference site clean?”
spyware / malware / malicious JSSandbox-loads any URL and reports a malware/phishing verdict, every script the page loads, the third-party domains it contacts, plus a screenshot — before you trust it as a template.
“Auto-block bad sites for the team”
always-on enforcementDNS + HTTP policies block known malware, phishing and spyware categories automatically for every user and device — no per-machine setup, full logging of what was blocked and why.
“Safely open an untrusted site”
zero local riskRenders the risky site in a remote sandbox at Cloudflare's edge. Malicious JS and drive-by malware run away from the user's machine — only safe pixels reach the endpoint.
“Is AI traffic using the governed path?”
real inference + observabilitySends a real clean model request through the configured AI Gateway and records the gateway, model, and log identifier.
“Does sensitive prompt data reach the model?”
synthetic block testSends fixed synthetic sensitive data and passes only when the AI Gateway Firewall DLP policy blocks it before inference.
“Bots scraping site or AI”
real users vs. machinesML scoring separates real users from automated traffic, blocks scraping of both the site and the AI endpoint, and controls AI crawlers so your content isn't harvested for training.
See the full experience before signing in: domain entry, all six controls, scan details, AI Gateway evidence, and a DLP block result.
Presentation-only recording using reserved example data. No request was sent to URL Scanner, AI Gateway, DLP, email, or the example domain.
One authenticated action runs Cloudflare URL Scanner, verifies a real AI Gateway route, and confirms that DLP blocks fixed synthetic sensitive data.
Ready. Enter a site and hit Scan.