Ismail HakimNovember 2026

Inside Hirudo: An Android Banking Trojan Built for Surveillance, Credential Theft, and Banking-App…

Inside Hirudo: An Android Banking Trojan Built for Surveillance, Credential Theft, and Banking-App Replacement

Cyberkarta Threat Research | November 14, 2026

Research category: Android Malware Analysis

Malware family: Hirudo

Platform: Android

TL;DR — Hirudo Malware Analysis

Hirudo is an Android banking trojan and Remote Access Trojan (RAT) designed to compromise financial applications through Accessibility abuse, credential theft, SMS interception, remote surveillance, and banking-application replacement.

Cyberkarta’s static-analysis research identified the following key findings:

  • Delivery: Malicious Android droppers, including KoiShort (com.koishort.app) and ZoyaTV (com.zoyatv.app), distribute Hirudo-family payloads through deceptive application-update workflows.
  • Payload execution: A native loader known as Crabshell (libmqttx.so) reconstructs and loads concealed DEX payloads while implementing anti-debugging and anti-instrumentation checks.
  • Command-and-control: Hirudo uses HTTPS for registration and data exchange, MQTT v5 for remote commands, and WebRTC components for surveillance. Its documented cryptographic mechanisms include ECDH, ECDSA, and AES-GCM.
  • Banking targets: The analyzed configuration contains 113 banking and payment application targets across six countries, including eight Indonesian banking applications associated with replacement workflows.
  • Credential theft: The malware includes functionality for collecting banking credentials, PIN-related information, financial SMS messages, and potentially OTPs.
  • Remote access and persistence: Documented components include screen capture, camera access, boot receivers, foreground services, and a native watchdog.

Indicators of Compromise (IoCs)

The following indicators were extracted from the archived static-analysis research.

File and application indicators

  • SHA-256 (Hirudo RAT): b2270c29b3d0a49a716730d6ef108cffec0b1a4b159dd02e93333fb85e492b99
  • RAT package: com.r9nkvx.qw2mpl
  • Application label: NKV Verify
  • KoiShort dropper: com.koishort.app
  • ZoyaTV dropper: com.zoyatv.app
  • Native loader: libmqttx.so
  • Loader marker: crabshell-nested-v2

Network indicators

  • C2 domain: www[.]helloharmonyos[.]com
  • C2 IP address: 134[.]122[.]153[.]136
  • MQTT endpoint: www[.]helloharmonyos[.]com:1883
  • Replacement APK infrastructure: d3q4tlm7pw29zr[.]cloudfront[.]net
  • SMS collection API path: /api/device/sms_batch

Detection opportunities: YARA signatures targeting distinctive loader artifacts, and Suricata rules monitoring associated C2 infrastructure, MQTT communication characteristics, and identifiable HTTP API patterns.

Research note: These findings are derived from archived static-analysis research. The listed network indicators are historical research artifacts; current infrastructure activity has not been independently verified.

Introduction

Android banking malware has evolved far beyond displaying fake login screens and stealing credentials.

Modern banking trojans increasingly combine remote access, accessibility abuse, encrypted communication, and sophisticated persistence mechanisms to take control of compromised devices.

Hirudo is an example of this evolution.

Based on Cyberkarta’s analysis of archived reverse-engineering research, Hirudo is a packed Android banking-fraud Remote Access Trojan (RAT) with functionality for collecting credentials, intercepting SMS messages, remotely interacting with banking applications, streaming device screens, and delivering modified versions of legitimate banking applications.

The analyzed malware contains a registry of 113 banking and payment application targets across six countries, including eight Indonesian banking applications with dedicated application-replacement workflows.

Unlike conventional banking trojans that primarily rely on phishing overlays, Hirudo combines accessibility-driven automation with a native payload loader, encrypted command-and-control infrastructure, and multiple persistence mechanisms.

Our investigation examines its delivery chain, internal architecture, command-and-control protocol, credential-theft capabilities, and implications for Android banking security.

Hirudo Infection Chain

1. How Hirudo Reaches Its Victims

The infection chain begins with Android applications that act as malware droppers.

Two applications examined in the research were:

  • KoiShort: com.koishort.app
  • ZoyaTV: com.zoyatv.app

Both applications share substantial code similarities and use a deceptive installation workflow.

Instead of immediately exposing malicious functionality, they display an interface resembling a Google Play update prompt.

The goal is to persuade the victim to install an additional Android application.

The droppers use Android’s PackageInstaller functionality and implement a VPN mechanism that can route traffic into a blackhole during the installation process.

However, their payload-packaging techniques differ.

ZoyaTV stores an embedded Android payload directly in assets/vpn.apk, while KoiShort uses additional encryption and obfuscation to conceal its embedded payload.

Both contain Hirudo-family payloads, although the available evidence does not establish that either delivered the exact RAT APK examined separately in this research.

Once installed, the RAT attempts to obtain sensitive Android permissions and services, particularly Accessibility access, notification access, overlay permissions, and VPN consent.

These capabilities form the foundation of its subsequent activities.

2. Crabshell: The Native Loader Hiding Hirudo’s Real Payload

One of Hirudo’s most interesting architectural characteristics is its native loader.

The analyzed RAT APK uses the package identifier com.r9nkvx.qw2mpl and application label NKV Verify.

Its visible Java bytecode does not expose the complete malware implementation.

Instead, the initial application behaves as a bootstrap component responsible for loading the actual payload.

The research identifies a native library called libmqttx.so containing a loader associated with the string crabshell-nested-v2.

We refer to this component as Crabshell.

Crabshell’s apparent responsibilities include reconstructing encrypted payload components, validating their integrity, loading DEX bytecode into memory, and replacing the application’s original Android Application object.

Several strings recovered during static analysis illustrate this behavior:

crabshell-nested-v2
load_dex_core
Payload reassembled shards=
Payload hash mismatch! Expected:
extract_assets_core
xurf_assets.zip

The loader also uses Android runtime manipulation to integrate the reconstructed application into the existing process.

This matters because conventional static analysis may initially reveal only the bootstrap layer rather than the malware’s complete implementation.

Anti-analysis mechanisms

Hirudo includes checks designed to interfere with reverse engineering.

The documented loader checks for active debugging sessions and instrumentation artifacts, including:

frida
xposed
lsposed
substrate

It also checks common Frida-related localhost ports, including 27042 and 27043.

When its anti-analysis checks detect suspicious conditions, the loader can terminate the application process.

The APK additionally contains deliberately inconsistent ZIP metadata, including misleading encryption flags and invalid CRC values.

These characteristics can interfere with standard APK parsing and decompilation utilities.

Research finding: The Crabshell loader and its anti-analysis checks are supported by archived static evidence. The exact native payload-reconstruction algorithm remains unresolved.

3. Encrypted Command-and-Control Architecture

Hirudo’s command-and-control architecture is more sophisticated than a simple HTTP polling mechanism.

The analyzed code contains two principal communication channels.

HTTPS: Used for device registration, enrollment, session initialization, telemetry, credential-related uploads, and retrieving operational configuration.

MQTT v5: Used for command delivery, execution reporting, device status, liveness, and WebRTC signaling.

The analyzed v202 configuration references:

HTTPS API:
<https://www.helloharmonyos.com>
MQTT broker:
tcp://www.helloharmonyos.com:1883

Other embedded payload versions reference the IP address 134.122.153.136 on different ports.

These addresses are historical research indicators and should not be interpreted as proof that the infrastructure is currently active.

Device authentication and encryption

Hirudo’s documented architecture uses an elliptic-curve P-256 key stored through AndroidKeyStore.

Requests can be authenticated using ECDSA signatures and headers such as:

X-Device-Id
X-Key-Id
X-Timestamp
X-Nonce
X-Signature
X-Session-Id

Its command protocol also incorporates ECDH-based key establishment and AES-GCM encryption.

This architecture provides confidentiality and authenticated command handling while incorporating mechanisms intended to reject replayed commands.

For defenders, this complicates simple payload-inspection approaches: malicious commands may be encrypted, and network traffic can resemble structured application communications.

That does not make Hirudo invisible.

Network defenders can still investigate destination infrastructure, MQTT connection characteristics, transport metadata, and correlations with suspicious endpoint behavior.

Why MQTT matters

MQTT is a lightweight publish-subscribe messaging protocol commonly used in IoT environments and applications requiring efficient communication.

Unlike traditional HTTP polling, MQTT allows a device to maintain a persistent connection with a message broker.

Commands and responses can be distributed through MQTT topics, allowing the attacker to communicate with compromised devices without repeatedly initiating separate HTTP requests.

For a remote-access trojan, this design offers several advantages:

  • Persistent communication: A compromised device can maintain an ongoing connection to its command infrastructure.
  • Asynchronous command delivery: Commands can be delivered through subscribed topics without waiting for a periodic HTTP polling interval.
  • Efficient communication: MQTT introduces relatively low protocol overhead.
  • Command separation: Different topics can organize commands, responses, device status, and signaling activity.
  • Encrypted application messages: Even when MQTT transport metadata is observable, command contents may remain inaccessible without the appropriate cryptographic material.

In Hirudo’s documented configuration, MQTT v5 is combined with encryption and authenticated device communication.

As a result, a network analyst may identify MQTT connections without being able to directly inspect the underlying malicious command payloads.

This makes endpoint telemetry, static artifacts, and network analysis valuable complementary detection sources.

4. Accessibility Abuse: The Core of Hirudo’s Banking Attacks

Android Accessibility Services are designed to help users interact with their devices.

However, when malicious applications obtain Accessibility access, the same capabilities can be abused to observe interfaces and perform interactions on behalf of the user.

Hirudo’s analyzed code uses accessibility events to identify which application is currently active.

Rather than relying primarily on a complete inventory of installed applications, the RAT monitors foreground application activity and matches package identifiers against its banking target registry.

Once a targeted banking application is detected, Hirudo can initiate application-specific handling.

Depending on the target and available capabilities, this may include:

  • Observing interface elements and accessibility events.
  • Capturing credentials or PIN-related input.
  • Automating navigation and interactions.
  • Displaying deceptive interfaces.
  • Triggering fake application-update workflows.

This makes Accessibility access particularly valuable to the attacker.

It potentially enables control over legitimate application interfaces without requiring root privileges.

The archived analysis did not identify a root exploit in the reviewed RAT.

5. Banking-App Replacement: A Particularly Dangerous Capability

One of Hirudo’s most significant features is its ability to initiate replacement workflows targeting genuine banking applications.

The research identified eight Indonesian banking applications associated with dedicated replacement paths.

Targeted Indonesian banking applications

  1. BRImo Package: id.co.bri.brimo
  2. Wondr by BNI Package: id.bni.wondr
  3. BSI Package: co.id.bankbsi.superapp
  4. BTN Mobile Package: id.co.btn.mobilebanking.android
  5. JakOne Mobile Package: com.dev.jakone.mbanking
  6. Livin’ by Mandiri Package: id.bmri.livin
  7. BCA mobile Package: com.bca
  8. myBCA Package: com.bca.mybca.omni.android

The malware contains paths pointing to replacement APK files hosted through a CloudFront distribution.

When a targeted application is detected, the RAT can display a deceptive update interface, retrieve a replacement APK, and attempt to automate the installation workflow.

It is important to distinguish this from silently modifying an installed banking application.

Standard Android package-signature enforcement still applies. A differently signed replacement generally cannot be installed as a normal update over the legitimate application. The attack may require an uninstall-and-reinstall sequence or other user-assisted installation steps.

The recovered research documents the download, interface, and installation logic, but does not include the actual modified banking APKs.

Therefore, the internals of those modified applications remain unverified.

What about YAHFA, Dobby, and Play Integrity bypasses?

Public reporting referenced in the archive describes additional techniques involving YAHFA, Dobby, modifications affecting FLAG_SECURE, and Play Integrity token relaying.

However, these mechanisms were not identified in the analyzed RAT implementation.

They appear to relate to a separate modified-banking-application stage, referred to in the research as stage C.

Until the stage-C APKs are obtained and independently analyzed, these mechanisms should be treated as publicly reported capabilities rather than confirmed findings from the available samples.

6. Credential Theft, SMS Interception, and Financial Surveillance

Hirudo also contains functionality for gathering sensitive information beyond conventional banking credentials.

The reviewed code includes application-specific credential storage routines, fake PIN interfaces, and upload mechanisms.

The malware can collect bank-related PIN information and other secrets obtained through its targeted workflows.

The research also documents SMS collection capabilities.

Its SMS scanning logic can examine previously received messages and identify messages associated with financial institutions or digital wallets.

Examples include:

  • OVO
  • GoPay
  • DANA
  • ShopeePay
  • LinkAja

The presence of these applications in the research does not necessarily indicate dedicated banking-app replacement functionality.

For the named digital wallets, the strongest documented evidence relates to financial SMS and potential OTP collection.

The reviewed SMS upload path includes:

/api/device/sms_batch

This capability is significant because SMS-based authentication remains present in some financial services and account-recovery workflows.

An attacker controlling both the victim’s device interface and accessible SMS messages may have opportunities to interfere with authentication and account recovery.

7. Remote Surveillance Through Screen and Camera Access

Hirudo’s functionality extends beyond credential collection.

The analyzed code contains components associated with Android MediaProjection, Camera2, and WebRTC.

These provide mechanisms for screen sharing, camera capture, and interactive remote communication.

The malware’s command surface includes references to screen and camera operations, alongside WebRTC signaling.

The research also identified a workflow for retrieving TURN credentials through its backend API.

Taken together, these components indicate an architecture designed to support remote device surveillance.

However, the actual reliability of screen streaming, camera access, and remote sessions on specific Android versions was not established through dynamic testing.

8. Persistence and Resistance to Removal

For banking-fraud malware, maintaining access to a compromised device can be as important as the initial infection.

Hirudo’s archived code analysis reveals multiple persistence-related mechanisms.

These include:

  • Boot receivers
  • Foreground services
  • Scheduled jobs
  • Background workers
  • Account synchronization components
  • Native watchdog functionality

The RAT separates functions across several Android processes, including processes associated with accessibility services, transport communication, and persistence.

It also contains defensive behavior intended to obstruct removal.

For example, accessibility automation can interfere when users navigate toward application-management interfaces associated with disabling or uninstalling applications.

These mechanisms can complicate user-driven removal.

Nevertheless, the presence of these routines does not prove that the malware can persist against every Android security mechanism or device configuration.

9. Understanding Hirudo’s Geographic Targeting

Hirudo’s target registry extends beyond Indonesia.

The archived research identifies 113 banking and payment application entries across six countries.

Geographic distribution

  • China: 34 target entries
  • Morocco: 30 target entries
  • Mexico: 29 target entries
  • India: 11 target entries
  • Indonesia: 8 target entries
  • Thailand: 1 target entry

Total: 113 target entries

The eight Indonesian targets contain dedicated application-replacement paths.

The remaining 105 entries are associated with credential or PIN-targeting logic, without equivalent replacement paths identified in the analyzed code.

The geographic distribution suggests that Hirudo was designed with a broader target registry than Indonesia alone.

However, a target registry does not establish actual infections in every listed country.

Determining the malware’s real-world geographic impact would require additional telemetry, incident-response evidence, or independently verified campaign reporting.

10. Detecting Hirudo: From Static Artifacts to Network Behavior

Hirudo demonstrates why malware detection benefits from combining several sources of evidence.

A single file hash, domain, or IP address can be useful during an investigation, but these indicators may become obsolete or produce false positives when used without context.

The archived research includes both YARA and Suricata detection artifacts.

File-based detection with YARA

YARA can help identify known Hirudo samples or distinctive components of the native loader.

Relevant characteristics include combinations of loader-specific strings, APK package metadata, and recognizable code artifacts.

The supplied research includes rules intended to recognize:

  • Hirudo RAT artifacts
  • Reconstructed DEX content
  • Crabshell loader
  • Native watchdog components

These signatures can help malware researchers classify suspicious Android applications and identify relationships between analyzed samples.

However, strings such as libmqttx.so or references to common Android libraries should not be treated as independently sufficient proof of infection.

Reliable detection requires distinctive patterns and appropriate rule conditions.

Network-based detection with Suricata

Suricata can identify observable network characteristics associated with Hirudo communications.

Potential detection opportunities include:

  • Known command-and-control destinations
  • MQTT connection patterns
  • Distinctive API paths
  • Recognizable HTTP request structures
  • Correlation between suspicious destinations and device activity

Static analysis can provide an initial basis for these rules.

For example, an API route or MQTT topic identified in code can form a detection hypothesis.

Dynamic analysis and captured network traffic are then valuable for determining whether the pattern actually appears during execution.

Static analysis and dynamic analysis as complementary evidence

Detection engineering does not necessarily require every indicator to originate from packet captures.

Static analysis can reveal configuration values, embedded endpoints, protocol strings, and communication structures that are suitable for initial Suricata rule development.

Dynamic analysis can verify whether those artifacts appear in actual traffic and identify characteristics that may not be visible in the application code.

A practical detection-development workflow includes:

  1. Identify suspicious artifacts through static reverse engineering.
  2. Extract candidate network indicators and protocol characteristics.
  3. Create preliminary Suricata signatures.
  4. Validate observable network behavior through controlled execution and PCAP analysis when available.
  5. Test signatures for false positives and compatibility with legitimate traffic.
  6. Refine and document the resulting rules.

This approach allows defenders to begin detection development from static evidence while maintaining a clear distinction between hypothesized and observed network behavior.

The indicators listed in the TL;DR provide a starting point for further investigation.

They should be correlated with additional evidence rather than used as standalone proof that a device is compromised.

11. What Hirudo Means for Mobile Banking Security

Hirudo illustrates a shift in the Android banking threat landscape.

The attack no longer depends exclusively on displaying a fraudulent login screen.

Instead, its documented architecture combines malicious application delivery, native payload concealment, accessibility-based interaction, encrypted communications, financial data collection, and mechanisms for delivering replacement banking applications.

For financial institutions, the implications go beyond protecting user credentials.

Security controls should also consider device integrity, application authenticity, runtime manipulation, anomalous transaction patterns, and the limitations of trusting security signals originating from a potentially compromised client.

Security considerations for financial institutions

Application integrity

Mobile banking applications should implement appropriate application-integrity checks, signing-certificate validation, and runtime security mechanisms.

These controls can increase the difficulty of application modification and impersonation, although they cannot guarantee protection against a fully compromised device environment.

Device trust

Device-integrity and attestation signals can contribute to fraud-risk assessment.

However, their results should not be treated as absolute proof that a banking session is secure.

Transaction monitoring

Financial institutions should correlate transaction behavior with device characteristics, authentication events, account history, and other relevant risk signals.

An authenticated transaction does not necessarily mean that the legitimate account holder intentionally initiated it.

Authentication security

Account protection should account for the possibility that SMS messages, notification contents, or screen interactions are accessible to malware on the same device.

Threat intelligence and detection

Indicators obtained through reverse engineering can support investigation, incident response, and the development of security detections.

However, detection content should be validated before being deployed broadly in production environments.

Security considerations for Android users

For Android users, installing applications from trusted sources, scrutinizing Accessibility and notification permissions, and avoiding unsolicited application-update prompts remain important preventive measures.

Users should also verify the authenticity of banking applications before installation and avoid approving unexpected requests to uninstall or replace financial applications.

Crucially, even a legitimate banking application cannot assume that the surrounding Android environment is trustworthy once a malicious application has gained extensive device-level permissions.

Conclusion

Hirudo is more than a conventional Android credential-stealing application.

Its documented architecture combines a native loader, encrypted command-and-control protocols, accessibility abuse, SMS collection, remote surveillance functionality, persistence mechanisms, and targeted banking-application replacement workflows.

The most significant finding is not simply the number of financial applications in its target registry.

It is the malware’s attempt to control the broader environment in which those applications operate.

Crabshell adds another layer of complexity by concealing the malware’s primary implementation behind native code, encrypted payload components, and anti-analysis mechanisms.

Meanwhile, Hirudo’s use of HTTPS, MQTT, and WebRTC demonstrates how multiple communication technologies can be combined within one Android malware architecture.

Its banking-application replacement functionality also highlights the risks of deceptive installation workflows and unauthorized application substitution.

For defenders, this reinforces an important principle:

Mobile banking security must account for both application-level vulnerabilities and threats originating from the surrounding device environment.

Understanding Hirudo’s architecture provides valuable insights into how modern Android banking trojans combine application abuse, device-level permissions, and encrypted remote communication to support financial fraud.

About Cyberkarta Threat Research

Cyberkarta conducts cybersecurity research, malware analysis, penetration testing, and incident-response investigations. Our research focuses on understanding real-world attack techniques and translating technical findings into actionable defensive insights.

Research source: Hirudo static-analysis archive, research dated November 10, 2026. Article prepared November 14, 2026.

Evidence terminology: Confirmed refers to direct static evidence documented in the archived research; Publicly Reported refers to external findings not independently verified in these samples; Inferred denotes analytical conclusions; Unverified identifies missing evidence.


Inside Hirudo: An Android Banking Trojan Built for Surveillance, Credential Theft, and Banking-App… was originally published in Cyberkarta on Medium, where people are continuing the conversation by highlighting and responding to this story.

Logo

PT Cyberkarta Tugu Teknologi

Platform belajar cyber security bersama komunitas cyber

Copyright © 2026 Cyberkarta