AI-Generated ClickFix Campaign Delivers SmartRAT Banking Trojan via Fake Brazilian Bank Website

Multiple instances of typosquatting domains hosting malicious content generated with AI-powered website creation tools. 

One striking campaign combined an AI-created fake Brazilian bank site with a ClickFix social-engineering lure to deliver a PowerShell-based remote access trojan Zscaler named SmartRAT.

The operation demonstrates how low-cost AI tooling reduces development friction for attackers while preserving sophisticated post-exploitation capabilities.

Threat actors registered a typosquatting domain (cartaobb[.]com) closely mimicking a legitimate bank domain and populated the site with high-fidelity visuals and templated code comments indicative of AI website generation.

The fraudulent page offered a credit-card application and staged a Cloudflare-like CAPTCHA.

When victims clicked the fake CAPTCHA, client-side scripts copied a PowerShell command into the clipboard, forced the browser into fullscreen, and displayed a fake Blue Screen of Death (BSOD) “system recovery” screen.

A lockdown routine disabled common DevTools shortcuts, cleared the console repeatedly, blocked most keystrokes while permitting only paste and Enter, and kept focus on the tab pressuring the user to paste and run the supplied command.

AI-Generated ClickFix Campaign

The clipboard PowerShell command fetched a stealth dropper (st.txt) from a hardcoded IP and executed it invisibly by hiding the PowerShell window via user32.dll.

In March 2026, Zscaler ThreatLabz documented an escalation in threat actors’ use of AI-powered website builders to generate convincing typosquatting pages at scale. 

If the window title matches a list of predefined targets, SmartRAT logs the title, matched keyword, process name, and timestamp, and sends this information to the SmartRAT C2 server as a BrowserAlert (message type 0x80).

The dropper saved a decoy file (msedge.txt) and executed it; msedge.txt then retrieved and decrypted an AES-CBC blob using hardcoded keys to produce the SmartRAT payload.

SmartRAT is implemented entirely in PowerShell and identifies itself with the embedded string SMART_V25.

SmartRAT’s architecture centers on encrypted C2, remote-control features, credential theft overlays, and robust persistence.

It decrypts two C2 endpoints (a domain resolved to c[.]windowsupdate-cdn[.]com and a fallback IP 162[.]141[.]111[.]227) and communicates over raw TCP on port 51888 using an AES-CBC framing scheme that transmits IV and ciphertext as hex-delimited pairs.

The RAT generates a unique machine token and computes an HMAC-SHA256 with a hardcoded master key; the same string is also used (after SHA-256) to derive the AES key. Debug and activity logs are written under ProgramData or AppData paths to conceal operations.

Operator capabilities are extensive: automatic screen capture and streaming, monitor enumeration with DPI-aware coordinates, a high-priority keylogger, clipboard theft, file browsing and exfiltration, and full-feature overlays that mimic bank-branded security prompts, Windows Update, or BSOD screens.

Notably, SmartRAT implements QR-code detection and QR-overlay swapping for bank flows that rely on QR authorizations an effective technique for intercepting or substituting payment confirmations.

The RAT supports commands to move the cursor, inject keystrokes, freeze and unfreeze input, display targeted overlays, and perform system actions such as restart or uninstall.

Persistence mechanisms include scheduled tasks and a registry Run key; if UAC elevation is granted, SmartRAT compiles C# service code and installs a Windows service to run as SYSTEM.

If elevation fails, it falls back to hidden PowerShell processes and scheduled tasks to reattempt elevation at logon. The payload compiles and uses C# helper components for screen capture, input control, and session token impersonation to dismiss lock screens.

Zscaler also identified an operational weakness: the SmartRAT web-based C2 panel contained client-side-only authentication checks relying solely on localStorage values (authToken and currentUser).

This trivial flaw allows an investigator or defender to bypass login by setting localStorage keys, exposing the panel’s administrative functions and underscoring poor operational security in AI-assisted development workflows.

Trend Micro independently reported the same family (calling it Banana RAT), documenting an alternate attack path prior to Zscaler’s blog.

This campaign highlights two converging trends: AI site builders let attackers rapidly generate convincing lures, and modular, script-based malware like SmartRAT delivers mature remote-control features without compiled binaries.

Defenders should monitor typosquatting registrations, enforce multi-layered detection for clipboard-pasted command executions, and prioritize network signatures for unusual TCP traffic on nonstandard ports and domain-to-IP fallbacks.

Indicators Of Compromise (IOCs)

IOC	Description
crefisa[.]online	Fraudulent domain
vfsgloball[.]net	Fraudulent domain
cartaobb.com	Fraudulent domain
windowsupdate-cdn[.]com	C2 domain
297eb45f028d44d750297d2f932b9c91	st.txt
6bf4d4c62b5138ace281ce3d08297787	payload[.]php
3c72e1f37f115b00c3ad6ed31bacfe8a	Powershell RAT
b17ccdb5531555e43f082d6e77c07227	Powershell RAT
64[.]95[.]13[.]238	C2 IP
162[.]141[.]111[.]227	C2 IP

Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Follow us on Google News, LinkedIn, and X to Get Instant Updates and Set GBH as a Preferred Source in Google.