SURXRAT is an actively developed Android remote access trojan (RAT) marketed as a malware-as-a-service offering under the “SURXRAT V5” branding and distributed through a Telegram-based ecosystem. Reporting cited in the content assesses it as likely operated by an Indonesian threat actor and indicates it likely evolved from the ArsinkRAT malware family, supported by code overlap and use of an "arsinkRAT" database reference. Researchers identified more than 180 related samples, with the Telegram marketing channel created in late 2024 and active development likely beginning in early 2025.
SURXRAT abuses Android Accessibility Services for persistent control and automation, and communicates with Firebase Realtime Database command-and-control infrastructure, including xrat-sisuriya-default-rtdb.firebaseio.com, which helps its traffic blend with legitimate cloud services. It registers infected devices using a random UUID and maintains a persistent background service.
Its capabilities include extensive surveillance, data theft, and remote device control. Reported collection includes contacts, SMS messages, call logs, Gmail account data, browser history, clipboard contents, location data, notifications, Wi-Fi history, cellular tower information, installed applications, files, device brand and model, Android version, battery status, SIM details, network information, and public IP address. Remote actions include recording audio, taking camera photos, sending SMS messages, placing phone calls, opening URLs, changing wallpapers, controlling the flashlight, vibrating the device, displaying toast or text-to-speech messages, unlocking devices, deleting files, uploading files, and wiping storage.
SURXRAT also includes a ransomware-style screen locker that can display a persistent full-screen lock message, enforce an attacker-defined PIN, report incorrect PIN attempts to the backend in real time, and be remotely removed by the attacker. The malware has been described as targeting Android devices broadly, with broader reporting placing it among threats actively targeting Brazilian PIX payment infrastructure.
A notable recent behavior is conditional download of a very large LLM module, reported as larger than 23 GB, from Hugging Face or other third-party repositories. This download is triggered when specific gaming apps are active, including Free Fire MAX x JUJUTSU KAISEN and Free Fire x JUJUTSU KAISEN, or when target package names are supplied remotely by the operator. The content states these AI modules can automate malicious tasks, including generating realistic phishing content, tailoring social-engineering prompts, and autonomously interacting with on-device apps and user interfaces to steal credentials or other sensitive data. Researchers also assessed the LLM-related behavior may be intended for device or network lag manipulation, masking malicious activity through degraded performance, evasion, future AI-assisted functionality, or alternative monetization.
8 distinct techniques documented for this family, organized by ATT&CK tactic.
"SurxRAT can generate realistic phishing content"
"tailor social-engineering prompts, and interact autonomously with on-device apps and user interfaces to exfiltrate credentials"
"Once installed, the apps urge users to enable accessibility services to realize their goals."; "TaxiSpy RAT... abuses Android's accessibility service..."; "The malware abuses accessibility permissions for persistent control"
"interact autonomously with on-device apps and user interfaces to exfiltrate credentials"
"exfiltrate credentials or sensitive data"
"interact autonomously with on-device apps and user interfaces to exfiltrate credentials"
"can download and run large language model (LLM) modules from third-party repositories"
"full-fledged remote administration tools such as SURXRAT"; "enabling threat actors to gather sensitive data and execute commands"; "comprehensive remote control"
"incorporate a ransomware-style screen locker module... deny access by displaying a full-screen lock message until a payment is made."
2 indicators attributed across vendor reports, sandbox runs, and researcher write-ups. Values are masked here and available in full in Mallory.
5 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.