Skip to navigation

Threat reportMalwareTL-2026-0142

SURXRAT Android RAT — LLM Module Downloads from Hugging Face, MaaS via Telegram, ArsinkRAT Evolution

highACTIVE

SURXRAT is an actively developed Android RAT commercially distributed via Telegram-based MaaS with structured reseller/partner licensing (200K-500K one-time payment). Built on ArsinkRAT code (1,216 APK hashes, 317 Firebase C2 endpoints, 45,000 victims across 143 countries), SURXRAT V5 operates as a full surveillance and device-control platform.

Key facts

IdentifiersTL-2026-0142
Also known asSURXRAT, SURXRAT V5, ArsinkRAT
SeverityHIGH
Affected productsGoogle Android
Weaknesses (CWE)CWE-200, CWE-306, CWE-940
AttributionSURXRAT Operator (medium confidence)
Nation-state nexusIndonesia
MalwareSURXRAT
MITRE ATT&CK techniquesT1414 Clipboard Data, T1420 File and Directory Discovery, T1422 System Network Configuration Discovery, T1426 System Information Discovery, T1429 Audio Capture, T1430 Location Tracking, T1437.001 Application Layer Protocol: Web Protocols, T1471 Data Encrypted for Impact, T1476 Deliver Malicious App via Other Means, T1481.002 Web Service: Bidirectional Communication, T1513 Screen Capture, T1516 Input Injection and 14 more in the attack flow
MotivationFINANCIAL
Target sectorsconsumer, financial, telecommunications, gaming
Target regionsMiddle East, Southeast Asia, Africa, South Asia, Europe, Americas
Detection rules9 (1 shown free)
Sources6 cited sources

Detection coverage

As of 2026-02-24, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0142 across Splunk SPL, Microsoft KQL and Sigma, covering 16 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: SURXRAT Full Surveillance Suite (SMS/Contacts/Calls/Location/Audio/Notifications), 23GB LLM Module from Hugging Face, and Ransomware Screen Locker

Detects post-exploitation: comprehensive Android surveillance — SMS monitoring and sending (OTP interception, fraud), contact/call log exfiltration, Gmail account theft, location tracking, notification interception (2FA tokens), clipboard monitoring, browser history, cellular tower intelligence, WiFi scanning, audio recording, camera capture, file enumeration. Conditional 23GB+ LLM download from Hugging Face triggered by Free Fire MAX/JUJUTSU KAISEN gaming apps or remote C2 command. Ransomware-style screen locker with attacker-defined PIN, customizable lock message, wrong PIN attempt logging to Firebase. Device Admin abuse for storage wipe.

Splunk SPL

index=proxy sourcetype="proxy_logs"
| search (
    (url_domain="xrat-sisuriya-default-rtdb.firebaseio.com" (http_method="PUT" OR http_method="PATCH"))
    OR (url_domain="*huggingface.co" bytes_out>5368709120)
    OR (url_domain="*googleapis.com" url_path="*script*macros*" http_method="POST")
    OR (url_domain="api.telegram.org" url_path="*bot*" http_method="POST")
    OR (url_domain="*firebasestorage.googleapis.com" http_method="POST")
)
| eval alert_type=case(
    match(url_domain, "xrat-sisuriya") AND (http_method="PUT" OR http_method="PATCH"), "surxrat_data_exfil_firebase",
    match(url_domain, "huggingface") AND bytes_out>5368709120, "surxrat_llm_module_download",
    match(url_path, "script.*macros"), "surxrat_apps_script_exfil",
    match(url_domain, "api.telegram.org") AND match(url_path, "bot"), "surxrat_telegram_bot_exfil",
    match(url_domain, "firebasestorage"), "surxrat_firebase_storage_upload",
    1=1, "surxrat_post_exploit"
)
| table _time, src_ip, url_domain, url_path, http_method, bytes_out, alert_type

Microsoft KQL

let firebase_exfil = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has "xrat-sisuriya-default-rtdb.firebaseio.com"
| extend AlertType = "surxrat_firebase_rtdb_exfil";
let llm_download = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has "huggingface.co"
| where SentBytes > 1073741824 or ReceivedBytes > 1073741824
| extend AlertType = "surxrat_llm_huggingface_download";
let apps_script = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has "script.google.com" or RemoteUrl has "googleapis.com/macros"
| where not(InitiatingProcessFileName has_any ("chrome", "msedge", "firefox", "sheets", "docs"))
| extend AlertType = "apps_script_exfil_non_browser";
let telegram_bot = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has "api.telegram.org" and RemoteUrl has "bot"
| where not(InitiatingProcessFileName has_any ("Telegram", "telegram"))
| extend AlertType = "telegram_bot_exfil_non_client";
let firebase_storage = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has "firebasestorage.googleapis.com"
| where not(InitiatingProcessFileName has_any ("chrome", "firebase"))
| extend AlertType = "firebase_storage_upload";
union firebase_exfil, llm_download, apps_script, telegram_bot, firebase_storage
| project Timestamp, DeviceName, AlertType, RemoteUrl, SentBytes, ReceivedBytes, InitiatingProcessFileName
| sort by Timestamp desc

Sigma

title: Large File Download from Hugging Face (>1GB) — SURXRAT LLM Module Indicator
id: 3a06a301-0fdd-46a5-a216-4b610f7a6564
status: experimental
description: Detects large file downloads (>1GB) from Hugging Face — SURXRAT conditionally downloads 23GB+ LLM module triggered by gaming app detection (Free Fire) or C2 command. LLM used for network latency manipulation, performance degradation masking, and AI-driven capability experimentation.
references:
    - https://cyble.com/blog/surxrat-downloads-large-llm-module-from-hugging-face/
author: Threadlinqs Team
date: 2026/02/24
tags:
    - attack.collection
    - attack.t1533
    - attack.t1471
    - attack.t1630.002
logsource:
    category: proxy
detection:
    selection:
        url.domain|contains: 'huggingface'
        bytes_out|gt: 1073741824
    condition: selection
falsepositives:
    - Legitimate ML/AI researchers downloading models from Hugging Face (verify device type — mobile downloads are suspicious)
level: high

Known false positives:

  • Legitimate ML researchers downloading Hugging Face models (verify mobile vs desktop)
  • Legitimate Firebase RTDB usage by enterprise apps

Tuning notes: Firebase RTDB PUT/PATCH from mobile = data exfiltration. 23GB+ Hugging Face download from mobile device is definitive SURXRAT LLM. Google Apps Script POST from non-browser = exfil via Google Drive. Telegram Bot API from non-Telegram client = bot-based exfil. 4 exfil channels: RTDB, Storage, Apps Script, Telegram Bot.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • SURXRAT Android RAT Multi-Signal Detection: Firebase C2 + Hugging Face LLM + Telegram Bot Exfil + Apps Script + Brand Impersonation APK · unlock with TL Creds
  • SURXRAT Credential Harvesting: OTP/SMS Interception, Notification Access (2FA Tokens), Clipboard Monitoring, and Gmail Account Theft · unlock with TL Creds
  • SURXRAT/ArsinkRAT Campaign IOCs: Firebase C2 Endpoint, Hugging Face LLM Source, Free Fire Gaming Triggers, and MaaS Distribution Channels · unlock with TL Creds
  • Android RAT Persistence Pattern: BOOT_COMPLETED Receiver, Foreground Service, Launcher Icon Suppression, and Accessibility Service Abuse · unlock with TL Creds
  • Android Comprehensive Surveillance: SMS Read+Send, Contact/Call Log Exfiltration, Location Tracking, Audio Recording, and Notification Interception · unlock with TL Creds
  • SURXRAT Ransomware Screen Locker: Device Admin PIN Lock, Wrong Attempt Logging to Firebase, Remote Storage Wipe, and Hybrid Monetization · unlock with TL Creds
  • SURXRAT MaaS APK Sideloading via Telegram/Discord/MediaFire, Brand Impersonation (50+ Brands), and BOOT_COMPLETED Persistence · unlock with TL Creds
  • SURXRAT Firebase RTDB C2 Infrastructure (14 Commands), 4 Exfiltration Channels, and Remote Device Control (Unlock/Wipe/Lock) · unlock with TL Creds

Indicators of compromise

As of 2026-02-24, Threadlinqs Intelligence tracks 16 indicators of compromise for TL-2026-0142.

Sample indicators (2 of 16)

  • Domain: xrat-sisuriya-default-rtdb[.]firebaseio[.]com
  • URL: hxxps://huggingface[.]co/

Defanged, TLP:CLEAR. Full set of 16 indicators: unlock with TL Creds.

16 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

Latest samples conditionally download a 23GB+ LLM module from Hugging Face, signaling experimentation with AI-assisted capabilities. Includes ransomware-style screen locker for device extortion.

SURXRAT is a next-generation Android Remote Access Trojan (RAT) that represents the convergence of mobile surveillance malware, AI experimentation, and commercial malware-as-a-service (MaaS) distribution. Uncovered by Cyble Research and Intelligence Labs (CRIL) in February 2026, SURXRAT has evolved from ArsinkRAT code into a feature-rich platform with novel capabilities.

The malware is marketed through a dedicated Telegram channel under the 'SURXRAT V5' branding, created in late 2024 with active development since early 2025. Over 180 related samples have been identified. The MaaS model offers two tiers: Reseller Plan (200K one-time, 3 builds/day, reselling rights) and Partner Plan (500K one-time, 10 builds/day, own reseller networks). As of January 2026, the platform reports 1,318 registered operator accounts.

SURXRAT evolved from ArsinkRAT, a cloud-native Android RAT previously tracked by Zimperium. ArsinkRAT used Firebase Realtime Database, Firebase Storage, Google Apps Script (uploading to Google Drive), and Telegram Bot API for C2/exfiltration. The predecessor campaign was massive: 1,216 unique APK hashes, 317 Firebase RTDB endpoints, and approximately 45,000 victim IPs across 143 countries. Largest victim concentrations in Egypt (~13K), Indonesia (~7K), Iraq (~3K), Yemen (~3K), and Türkiye (~2K). Distributed via Telegram, Discord, MediaFire links, impersonating 50+ brands (Google, YouTube, WhatsApp, Instagram, TikTok).

SURXRAT's key innovation is the conditional download of a large LLM module (>23GB) from Hugging Face. This download triggers when specific gaming applications are active (Free Fire MAX, Free Fire x JUJUTSU KAISEN) or when the C2 server sends target package names dynamically. The LLM module serves multiple purposes: deliberately introducing network latency during gameplay (paid cheating/disruption services), masking malicious background activity by degrading device performance, and enabling future AI-driven capabilities (automated interactions, adaptive social engineering).

The malware's C2 infrastructure uses Firebase Realtime Database (hxxps://xrat-sisuriya-default-rtdb.firebaseio[.]com) with the database reference labeled 'arsinkRAT', confirming the code lineage. Device registration uses random UUID for victim tracking.

SURXRAT provides comprehensive surveillance: SMS monitoring, contact/call log collection, Gmail account data, location tracking, notification interception, clipboard monitoring, browser history, cellular tower intelligence, WiFi scanning, file manager access. Remote control includes: device unlock, phone call initiation, wallpaper modification, audio playback, push notifications, forced URL opening, flashlight/vibration control, on-screen text overlays, and storage wipe.

The ransomware-style screen locker enables device locking with attacker-defined PIN and customizable lock message. Wrong PIN attempts are logged and transmitted to the operator in real-time. The lock can be remotely removed, giving complete control over device availability. This enables hybrid monetization: surveillance, fraud, or direct extortion based on victim value.

The Indonesian threat actor operates the platform, targeting aspiring cybercriminals rather than conducting attacks directly. The structured pricing, operational announcements, and feature updates demonstrate mature commercialization similar to underground SaaS platforms.

MITRE ATT&CK techniques by tactic:

Credential Access

T1414 Clipboard Data; T1517 Access Notifications

Discovery

T1420 File and Directory Discovery; T1422 System Network Configuration Discovery; T1426 System Information Discovery; T1430 Location Tracking

Collection

T1429 Audio Capture; T1513 Screen Capture; T1533 Data from Local System; T1636.002 Protected User Data: Call Log; T1636.003 Protected User Data: Contact List; T1636.004 Protected User Data: SMS Messages

Command and Control

T1437.001 Application Layer Protocol: Web Protocols; T1481.002 Web Service: Bidirectional Communication

Impact

T1471 Data Encrypted for Impact; T1643 Generate Traffic from Victim

Initial Access

T1476 Deliver Malicious App via Other Means; T1660 Phishing

Defense Evasion

T1516 Input Injection; T1628.001 Hide Artifacts: Suppress Application Icon

Persistence

T1541 Foreground Persistence; T1624.001 Event Triggered Execution: Broadcast Receivers

Execution

T1575 Native API

defense-evasion

T1630.002 File Deletion

Exfiltration

T1639 Exfiltration Over Alternative Protocol; T1646 Exfiltration Over C2 Channel

Remediation

Immediate actions

  • Block Firebase RTDB endpoint: xrat-sisuriya-default-rtdb.firebaseio.com
  • Hunt for SURXRAT/ArsinkRAT APK hashes on managed Android devices
  • Monitor for Hugging Face large model downloads (>23GB) from mobile devices
  • Block known distribution channels on Telegram for SURXRAT V5
  • Audit Android devices for suspicious accessibility service permissions

Workarounds

  • Restrict sideloading on enterprise Android devices via MDM policy
  • Block accessibility service grants for non-approved applications
  • Monitor for unusual large downloads (>1GB) on mobile devices

Longer-term hardening

  • Deploy mobile threat defense (MTD) with behavioral detection for Android RATs
  • Implement app vetting for sideloaded APKs on enterprise Android devices
  • Monitor Firebase RTDB traffic patterns for C2 communication indicators
  • Enable Google Play Protect on all managed Android devices
  • Educate users about risks of sideloading APKs from Telegram/Discord/MediaFire

Affected and fixed versions

  • Google — Android
    Vulnerable versions: All versions supporting accessibility services

Timeline

  • SURXRAT operator creates Telegram channel for MaaS distribution. Suggests active malware development began shortly after.
  • SURXRAT active development begins, building on ArsinkRAT codebase. Indonesian TA establishes MaaS platform with reseller/partner licensing.
  • Zimperium reports increase in ArsinkRAT activity targeting Android devices. 1,216 unique APK hashes, 317 Firebase endpoints, 45,000 victim IPs across 143 countries.
  • Zimperium coordinates with Google to take down malicious Firebase RTDB endpoints and Google Apps Script accounts used by ArsinkRAT. Google confirms known variants not on Google Play.
  • Zimperium publishes comprehensive ArsinkRAT analysis documenting 4 operational variants, cloud-native C2 infrastructure, and global victim distribution. Source: https://zimperium.com/blog/the-rise-of-arsink-rat
  • SURXRAT V5 branding appears. Operator publishes stats: 1,318 registered accounts, bot status active. Structured reseller (200K) and partner (500K) pricing established.
  • Cyble identifies SURXRAT samples conditionally downloading 23GB+ LLM module from Hugging Face. Download triggered by gaming app detection (Free Fire) or remote C2 commands.
  • Cyble publishes detailed SURXRAT analysis documenting AI-augmented capabilities, MaaS ecosystem, ransomware-style screen locker, and Firebase C2 infrastructure. Source: https://cyble.com/blog/surxrat-downloads-large-llm-module-from-hugging-face/
  • As of 2026-05-29, SURXRAT remains an actively developed Telegram MaaS Android RAT with no CVE, no arrest, and no takedown reported (Cyble Feb 2026, Gurucul Mar 2026); operator stats show "Bot Status: Active" with 1,318 accounts and live Firebase C2. It survived ArsinkRAT's partial June-2025 takedown by rebuilding, and ongoing Hugging Face LLM experimentation confirms active capability expansion.

Sources cited

Previous and next in malware

Drafted by the Threadlinqs research pipeline from 6 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats