Candiru

Overview

Candiru (operating under corporate entities Saito Tech Ltd, Taveta Ltd, and Grindavik Solutions) is an Israeli commercial cyber-surveillance contractor specializing in advanced zero-day mercenary spyware targeting Microsoft Windows, with modular extensions for macOS, iOS, and Android. Documented and forensically analyzed in July 2021 by The Citizen Lab and Microsoft Threat Intelligence Center (MSTIC), Candiru's primary espionage implant—tracked by Microsoft as 'DevilsTongue' and internally codenamed 'Caramel' or 'Sherlock'—is weaponized by government customers to execute covert surveillance against high-profile journalists, human rights activists, legal counsel, and political opposition figures across Europe, the Middle East, and Latin America. Candiru specializes in zero-click and precision one-click exploit chains: on Windows endpoints, it weaponized browser remote code execution zero-days (such as the Google Chrome WebRTC flaw CVE-2022-2294 discovered by Avast) chained with twin Windows GDI/win32k kernel privilege escalation zero-days (CVE-2021-31979 and CVE-2021-33771) to escape sandboxes and establish SYSTEM-level execution. Once deployed, DevilsTongue acts as an invasive multi-modular surveillance framework: it hooks process memory to extract end-to-end encrypted chat logs (Signal, WhatsApp, Telegram, Skype), harvests saved browser passwords and active session cookies to bypass Multi-Factor Authentication (MFA), exfiltrates Outlook message spools, captures microphone audio and keystrokes, and executes LSASS memory dumping to harvest Kerberos and NTLM credentials for network lateral movement.

Known aliases

Threat reports may refer to this family under multiple names:

• Candiru Spyware
• DevilsTongue
• Saito Tech
• KANDIRU
• Caramel Spyware
• Taveta
• Grindavik Solutions
• Sherlock (Internal Code)

Threat Actor Attribution

Attribution is probabilistic, not a science. Each link below states its confidence level and the evidence behind it.

• Candiru / Saito Tech Ltd High confidence
  Basis:

Family Pedigree & Lineage

Malware authors regularly fork and iterate upon existing codebases. Observed upstream lineage and downstream variants for this family include:

Derived From / Upstream Ancestry: Candiru Surveillance Framework

Known Derivatives & Downstream Variants: Devilstongue_Windows · Caramel_Loader · Candiru_Ios · Candiru_Android

MITRE ATT&CK Techniques

Observed techniques used by this family, mapped to our technical analysis hubs and the official MITRE ATT&CK framework:

Tactical Mitigations

Based on the techniques used by this family, consider the following defensive strategies:

• T1003.001: Monitor for LSASS memory dumping or registry SAM extraction. Enable Credential Guard on Windows systems.
• T1056.001: Implement Endpoint Detection and Response (EDR) to monitor for suspicious API calls related to keystroke interception. Enforce Multi-Factor Authentication (MFA) to render stolen passwords useless.
• T1071.001: Implement web filtering and SSL/TLS inspection to detect malicious command and control (C2) traffic hiding in HTTP/HTTPS.
• T1113: Monitor for frequent or unauthorized screen captures or clipboard reads. Restrict background applications from accessing desktop APIs.

MITRE D3FEND™ Defensive Countermeasures

Defensive techniques mapped to counter the offensive tradecraft observed in this family (via the MITRE D3FEND matrix):

Exploited Vulnerabilities (CVEs)

This malware family is documented weaponizing the following common vulnerabilities and exposures:

• CVE-2021-31979 — In-depth exploitation analysis and catalog intelligence (NIST NVD ↗).
• CVE-2021-33771 — In-depth exploitation analysis and catalog intelligence (NIST NVD ↗).
• CVE-2022-2294 — In-depth exploitation analysis and catalog intelligence (NIST NVD ↗).
• CVE-2021-21166 — Vulnerability documentation and metrics (NIST NVD ↗).
• CVE-2021-30551 — Vulnerability documentation and metrics (NIST NVD ↗).

Command & Control (C2) and Network Profile

How this family talks to its operators. C2 traffic can often identify a family even when its code and configuration are hard to analyse. Ports and patterns are historical and change between campaigns, so use them as hunting leads rather than block-only rules.

Indicators of Compromise

Public indicators drawn from CISA advisories and vendor reporting. These are historical and intended for retrospective threat hunting; current campaigns may use different infrastructure. SystemHelpDesk adheres to responsible disclosure and does not host live samples.

File hashes (SHA-256, sample)

Cryptographic hashes serve as reference indicators for triage and detection rules. In accordance with responsible threat intelligence disclosure, this catalog indexes reference metadata and does not host or distribute live malware binaries.

• 4e311f59223755ba9e74403cd3b036fa7c617a2fbcfcbb2e0f2f35fe99f927b2
• 92e10db56306e00b3e83b8b1784fbe3efaa447a123f1e917d23d9b0151a66e4a
• a3b1eb887f43398be3bfbb43952baab464010a30bfeee37c68ba02a6ce2bbffb

Delivery vectors

• Zero-click Chrome WebRTC heap buffer overflow exploitation (CVE-2022-2294)
• Watering-hole URL redirection targeting regional news, diplomacy, and civil society sites
• Spear-phishing delivery leveraging shortened URLs mimicking media and NGO domains

Persistence mechanisms

• Process injection into critical Windows daemons (svchost.exe, lsass.exe)
• Exploitation of Windows GDI kernel drivers (CVE-2021-31979, CVE-2021-33771) for ring-0 elevation
• Execution of custom signed/unsigned kernel drivers with anti-forensic log tampering

Tiered Detection Rules (Cisco CAIRN Model)

Modern threat detection requires multi-layered rule engineering. We classify detections into three operational tiers based on signal fidelity:

rule MALWARE_WIN_CANDIRU_T1_ARTIFACT {
    meta:
        description = "Tier 1 Primitive Artifact: Raw string matches for Candiru"
        tier = "T1 - Primitive Artifact"
        confidence = "Low-Medium (Fast Triage)"
        author = "SystemHelpdesk Threat Research Desk"
        date = "2026-10-14"
    strings:
        $s1 = "candiru" ascii wide nocase
        $s2 = "candiru spyware" ascii wide nocase
        $s3 = "devilstongue" ascii wide nocase
        $s4 = "saito tech" ascii wide nocase
        $s5 = "kandiru" ascii wide nocase
        $s6 = "caramel spyware" ascii wide nocase
        $s7 = "taveta" ascii wide nocase
        $s8 = "grindavik solutions" ascii wide nocase
        $s9 = "sherlock (internal code)" ascii wide nocase
    condition:
        uint16(0) == 0x5a4d and any of them
}

title: SystemHelpDesk Detection - Candiru (Spyware) Execution & Triage
id: a7d9aca0-7e98-581c-98fa-0ec5f4cf4a45
status: production
description: Detects execution, defense evasion, and persistence behaviors associated
  with the Candiru malware family. Detects suspicious execution from temporary user
  directories with hidden window flags and background transfer utilities.
references:
- https://malware-families-catalog.systemhelpdesk.com/families/candiru
- https://malware-families-catalog.systemhelpdesk.com/defensive/ransomware-protection
- https://malware-families-catalog.systemhelpdesk.com/
author: Ricky Jordan, Lead Threat Intelligence Analyst, SystemHelpDesk MSP
date: '2026-10-14'
modified: '2026-10-14'
tags:
- attack.spyware
- author.ricky-jordan
- source.systemhelpdesk-mfc
- attack.t1203
- attack.t1068
- attack.t1055
- attack.t1056_001
- attack.t1113
logsource:
  category: process_creation
  product: windows
detection:
  selection_process:
    Image|endswith:
    - \cmd.exe
    - \powershell.exe
    - \cscript.exe
    - \wscript.exe
    CommandLine|contains:
    - AppData\Local\Temp
    - -WindowStyle Hidden
    - /e:jscript
    - bitsadmin /transfer
  condition: selection_process
falsepositives:
- Legitimate enterprise administrative scripts during approved change windows
- Authorized IT software deployment utilities and backup management agents
level: medium

rule MALWARE_WIN_CANDIRU_T3_ATTRIBUTION {
    meta:
        description = "Tier 3 Attribution Fingerprint: Heuristic detection for Candiru tradecraft"
        tier = "T3 - Attribution Fingerprint"
        confidence = "High (Confirmed Family Signature)"
        author = "SystemHelpdesk Threat Research Desk"
        date = "2026-10-14"
    strings:
        // Core family identifier & heuristic payload signatures
        $attrib_str = "candiru" ascii wide nocase
        $pe_magic = "MZ" ascii
        $net_marker = "http" ascii nocase
    condition:
        uint16(0) == 0x5a4d and ($attrib_str and $net_marker) and filesize < 15MB
}

Containment & Response Steps

Ordered checklist for responders. Adapt to your environment and engage professional support for active incidents.

1. Sever all network connectivity immediately (disconnect physical Ethernet and disable all wireless/cellular interfaces) to prevent real-time C2 exfiltration.
2. Perform immediate volatile memory capture using specialized forensic hardware or live triaging kits before powering down or rebooting.
3. Enforce enterprise-wide session revocation: invalidate all OAuth tokens, active session cookies, and Kerberos tickets, and initiate full credential resets from a known-secure terminal.
4. Conduct bare-metal reimaging and firmware verification on all suspected endpoints, as DevilsTongue installs deep kernel hooks that evade standard operating system uninstallation.

What to Avoid

Common mistakes during response to this family that can destroy evidence, spread the infection, or worsen recovery.

1. Do not trust standard user-mode antivirus or endpoint scans to confirm eradication, as DevilsTongue hooks kernel memory structures and actively masks its footprint.
2. Avoid accessing corporate cloud repositories, webmail, or encrypted chat sessions on a suspect host until forensic validation is complete.
3. Do not assume that restarting the system neutralizes the threat, as kernel driver hooks and persistent scheduled tasks survive reboot cycles.

References & External Analysis

• The Citizen Lab: Hooking Candiru - Another Mercenary Spyware Vendor Comes into the Light
• Microsoft MSTIC: Protecting customers from a private-sector offensive actor using 0-day exploits and DevilsTongue malware
• Avast Threat Intelligence: The Return of Candiru - Zero-day in Google Chrome Used to Target Journalists
• CISA Known Exploited Vulnerabilities Catalog: CVE-2021-31979, CVE-2021-33771, CVE-2022-2294
• CISA Advisory: CISA Known Exploited Vulnerabilities (KEV) Catalog: CVE-2021-31979, CVE-2021-33771, CVE-2022-2294
• Search "candiru" on VirusTotal (External Analysis)
• Search "candiru" on Any.Run (Sandbox Sandbox)
• Search "candiru" on MalwareBazaar (Malware Samples)

North American Authority Advisory (CISA / NIST)

The Cybersecurity and Infrastructure Security Agency (CISA) has published an official US federal advisory on this family, which can be mapped to NIST Cybersecurity Framework (CSF) incident response controls: CISA Known Exploited Vulnerabilities (KEV) Catalog: CVE-2021-31979, CVE-2021-33771, CVE-2022-2294 ↗

Frequently Asked Questions

Related Threats (Category: Spyware)

Explore other documented threats in the same operational category:

• Predator
• Fearless_Lite
• Qqlogger
• Fearless_Key_Spy
• Relevantknowledge
• Hookey
• Formbook
• Pegasus

Machine-readable

Get this profile as JSON: https://malware-families-catalog.systemhelpdesk.com/api/candiru.json

Ecosystem & Interactive Environments

This profile is part of the Malware Families Catalog, a public dataset of 2,899 malware families. The catalog is also published across our ecosystem: Hugging Face, Kaggle, Zenodo, Replit, StackBlitz, CodeSandbox, and CodePen.