Edge Case

Brevo, a customer relationship management and digital marketing platform, confirmed a supply-chain attack on September 14 that injected malicious ClickFix scripts into its websites and JavaScript files embedded on customer sites. The attack lasted approximately five and a half hours, from 16:07 to 20:30 UTC.[1]

The root cause is a familiar story. Attackers obtained a long-lived Cloudflare API key with full account permissions that had been hardcoded in application source code. They used it to create a malicious Cloudflare Worker that modified content at the CDN edge, rewriting responses and stripping security headers including Content-Security-Policy. Because the Worker operated at the edge, Brevo's origin servers and files remained unmodified, and standard integrity checks detected nothing.[1]

The affected pages included brevo.com, sendinblue.com, login and onboarding pages, and sibforms.com. The Worker also modified the Brevo forms script, the Brevo Conversations widget, and the Brevo SDK loader scripts that customers embed on their own websites. Any site using those embeds was serving the injected ClickFix payload during the attack window.[1]

Brevo says the key may have been compromised as early as late August, though there is no evidence of prior malicious activity. Upon detection, the company revoked the key, removed the Worker and its routes, deleted attacker-controlled hostnames, and purged edge caches. App.brevo.com, the API, email delivery infrastructure, and customer account data were not affected.[1]

The lesson is not new, but it keeps being ignored. Hardcoding API keys in source code is a supply-chain attack waiting to happen. A long-lived key with full account permissions is the worst kind. And edge-level modifications that strip security headers and rewrite responses are invisible to origin-side integrity checks. If your CDN can serve modified content without your origin knowing, your integrity checks need to cover the edge too.[2]

In the same 24 hours, Zimperium disclosed a new Android malware called RatHat that uses an AI-powered subsystem to help operators remotely navigate compromised devices. The malware abuses Accessibility permissions, enables Wireless Debugging to gain shell-level execution, and installs a Go-based agent that restores itself if removed. The LLM prompts in the malware were written in Chinese, leading researchers to link it to threat actors from China. It is distributed through malvertising, SMS, and phishing sites promoting APK downloads from outside Google Play.[3]

An AI-powered Android banking trojan and a CDN-edge supply-chain attack in the same news cycle. One targets the endpoint, the other targets the infrastructure. Both rely on the same assumption: that the layer between the user and the real thing, the accessibility service, the CDN Worker, the API key in source code, is trusted by default. It is not.[2]