01 · High-speed · agentless

Sandbox analysisObserves OS and app behavior at high speed

Our own lightweight hypervisor rapidly observes behavior across multiple OS and app versions.
With no agent inside the VM, it resists sandbox detection and sleep-based bypasses.

BEHAVIOR · PLAYBOOK · AI VERDICT

Behavior extraction and a behavior playbook
for AI verdicts on malicious behavior

The Hyper Sandbox runs the document to extract events and behaviors, then chains the behaviors and matches them against attack sequences.
In this sample, the shell → remote download → executable creation → execution sequence matched, and the AI judged it Risk 10 · malicious.

File name, events, scores, techniques and URL: example values from a sample analysis report (analysis flow)

CORE TECH · HYPER SANDBOX

Technology elements of Hyper Sandbox

In a decoy Office, macro detection, deobfuscation and vulnerability traps expose hidden malicious code.
It is built on three technologies: Hyper Analysis, Forward Engineering and Morphing Sandbox.

Hyper SandboxUltra-fast deep malware analysis

3 TECH

  1. Hyper Analysis1 Hyper analysis

    In an environment optimized for analysis,
    fast and light · solves fragmentation

    Decoy Office / lightweight hypervisor
    • Office 2017 vulnerability
    • Office 2018 vulnerability
    • …
    • Office 2023 vulnerability
    • Remote file downloader
  2. Forward Engineering2 Forward engineering

    Forces button and hyperlink clicks
    to run for fast detection

    EVENTEVENTEVENT
  3. Morphing Sandbox3 Morphing sandbox

    An environment that changes every run
    neutralizes sandbox evasion

    Virtual Engine
CORE TECH · NEXT-GEN DETECTION

Next-gen advanced threat detection

Microservice architecture (MSA) · many small engines come together as one analysis engine.

Splitting the work by role keeps complexity low and makes maintenance and scaling easy.
At run time it reads the file's characteristics and reconfigures the engines in real time for the best result.

KISA · Korea Internet & Security Agency2025 excellent information security technologyLightweight hypervisor-based
UNKNOWN malware detection
Highly evolved advanced threat types
  • 01
    Office vulnerabilitiesExploits · remote template injection (RTI) · macros
  • 02
    Remote server URLsWeb servers · remote template injection · C&C servers
  • 03
    ScriptsVBS · PowerShell (PS) · JavaScript (JS) · batch files (BAT)
  • 04
    ObfuscationBinary HEX encoding · BASE64 encoding · custom encoding
  • 05
    Anti-detectionDeliberate time delays · random loop-escape puzzles · lying dormant until the next boot
Observed directly on our own lightweight hypervisor
DECOY
Decoy
Office
Decoy
browser
Decoy
interpreter
TRAP
Exploit
detection trap
URL
extraction trap
Deobfuscation
trap
Anti-evasion
trap
ENGINE
Threat element
extraction engine
URL
analysis engine
Script behavior
extraction engine
ENGINE
Interpretation engine
Malware verdict
AI engine
Verdict
Malware
detected & blocked

Ransomware · infostealers · backdoors · cryptominers and more

Analysis flow

Known threats fast,
unknown threats deep

Known threats are screened out by the pre-filter first; only the remaining files get deep analysis in the Hyper Sandbox.
AI maps behavior to MITRE ATT&CK techniques, scores the risk and hands it to SOAR · SIEM as report.json.

ZX CORE ENGINE

The ZX CORE engine in five parts

ZX CORE is a lightweight, fast sandbox analysis engine that handles static, dynamic and AI analysis in one engine.
Its name comes from Zeta(x): analyzing a file (x) with a sandbox (Zeta).

01

ZX CORE engine architecture

6-layer architecture · under 1 s analysis · 1 million samples a day

LAYERED ARCHITECTURE

Six layers produce a single verdict

Each analysis stage runs on its own, and the results are merged into one.
It is split into six layers: input → static analysis → dynamic analysis → interpretation → output → integration.

  1. Layer 01InputFile Intake · API GatewayReceives files for analysis and preprocesses them
  2. Layer 02Static analysisSAMSignature analysis · decryption · deobfuscation · string extraction
  3. Layer 03Dynamic analysisHyper Sandbox + DBAMRuns files in an isolated environment and monitors behavior
  4. Layer 04InterpretationTCE · AI classification engineClassifies and interprets malware with machine learning
  5. Layer 05OutputReporting ModuleGenerates structured analysis reports
  6. Layer 06IntegrationAPI · SIEM ConnectorIntegration with external platforms · IOC sharingPlanned
DATA FLOW

How a single file gets its verdict

Files received via REST API are statically analyzed, then run in the Hyper Sandbox to collect their behavior.
AI classifies the malware family and threat level, and the result is output as a JSON report.

  1. 1ReceiveReceives files for analysis via REST API
  2. 2Static analysisIdentifies the file type and removes encryption and obfuscation
  3. 3Sandbox allocationRuns the file in an isolated Hyper Sandbox
  4. 4Behavior collectionCollects API calls, network, registry and file changes
  5. 5AI analysisClassifies the malware family and threat level
  6. 6ReportOutputs the result in JSON format
PERFORMANCE

Performance specifications

In the default configuration, 8 instances analyze in parallel and each sample takes under 1 second.
It handles 1 million samples a day, and with an in-house hypervisor and Office, no separate licenses are needed.

  • Concurrent analysis instances8Default configuration
  • Analysis time per sampleUnder 1 sAvg. 0.856 s
  • Max file size256 MBConfigurable
  • Daily capacity1 millionsamples
  • Snapshot restore timeUnder 1 s
  • Licenses requiredNoneIn-house hypervisor · Office

Values are based on technical specification v1.4. The 0.27 s file analysis time from the certified test (KBeT) is shown under "Certifications & patents".

02

Transparent hypervisor execution

VM trace removal · Office vulnerability traps

TRANSPARENT EXECUTION

Malware never notices the virtual environment

It runs on our own hypervisor and removes the traces (artifacts) that reveal a virtual environment.
Here is how the environment malware sees differs between a typical VM and ZX CORE.

What malware sees · typical VM
  • CPU info and timing values reveal virtualization
  • VM registry keys and processes are visible
  • Virtual device hardware information
  • The same user name and paths every time

→ It recognizes the analysis environment and hides

ZX CORE Hyper Sandbox
  • Normalizes CPUID and RDTSC results
  • Masks VM-detection registry keys and process lists
  • Emulates hardware info to match real physical devices
  • Randomly generates user names and file paths

→ It sees a real PC and reveals its malicious behavior

OFFICE · SCRIPT RUNTIME

Diverse Office and script runtimes

It includes Office vulnerability traps for multiple versions and script emulators.
Documents actually trigger the vulnerabilities they target, and the URLs and C2 IPs they try to reach are extracted.

  1. 01
    Office vulnerability traps

    With traps for multiple Office versions, documents actually trigger the vulnerabilities they target.

  2. 02
    Script emulator

    Removes obfuscation and extracts behavior command by command.

  3. 03
    URL · C2 IP extraction

    Extracts the URLs and command server (C2) IPs a file tries to reach while running.

03

Static analysis module

Decompression · decryption · deobfuscation · string decoding

STATIC ANALYSIS · SAM

We look inside before anything runs

Without running the sample, it analyzes file structure, code and metadata for an initial threat assessment.
It unpacks, decrypts and deobfuscates files, then extracts and decodes encoded strings.

  1. 01
    Decompression

    Unpacks archives and analyzes the files inside.

  2. 02
    Decryption

    Decrypts encrypted MS Office files (some known methods).

  3. 03
    Deobfuscation

    Reveals code hidden with comments and line-joining or line-splitting syntax.

  4. 04
    String extraction

    Extracts ASCII and Unicode strings, and automatically decodes strings encoded with Base64, XOR and more.

SUPPORTED FORMATS

Formats supported by static analysis

Static analysis covers 10 document, 3 script, 1 shortcut and 15 archive formats.

Documents10DOCDOCXXLSXLSXPPTPPTXPDFRTFONEHWP
Scripts3JSVBSBAT
Shortcuts1LNK
Archives157ZBZ2CABGZISOLZ4LZHRARTARTBZXZZIPZSTALZEGG
04

Dynamic behavior analysis module

5 anti-evasion techniques that draw out hiding malware

ANTI-EVASION

Five techniques that draw out hiding malware

Malware that sleeps for a long time or waits for a click is drawn out within the analysis window.
The five are time acceleration, user action simulation, environment adaptation, multi-execution triggers and random puzzle escape.

  1. 01
    Time acceleration

    Manipulates the internal clock to wake malware that sleeps long before acting, within the analysis window.

  2. 02
    User action simulation

    Finds and clicks every hidden button and link.

  3. 03
    Environment adaptation

    Changes fixed values such as user name, IP, MAC, CPUID and SystemUUID on every run.

  4. 04
    Multi-execution triggers

    Fires multiple events, such as buttons and links, at the same time.

  5. 05
    Random puzzle escapePatent

    Breaks out of infinite loops that use random values to stall analysis.

05

AI threat classification engine

0–10 score · 5 levels · 6 families

THREAT SCORE

A 0–10 score in five levels

A machine learning model combines static and dynamic analysis results to classify malware and assign a threat score.
There are five levels in all, from 8–10 points for CRITICAL (malicious) to 0–1 points for BENIGN.

  1. CRITICAL8 ~ 10Malicious detected
  2. HIGH6 ~ 7Potentially malicious points detected
  3. MEDIUM4 ~ 5Suspicious · needs deeper analysis
  4. LOW2 ~ 3Meaningful logs · information
  5. BENIGN0 ~ 1Logs · information
MALWARE FAMILIES

Six malware families classified by behavior

Malware is grouped into six families based on the behavior it shows while running.
Fileless · process execution · user execution lures · auto-run · worm · bot · script execution.

  • FilelessFilelessProcessInjectionShellCodeInjectionetc.
  • ExecuterProcess executionDropFileExecDownloadFileExecetc.
  • FakerUser execution luresExeFileFakerSysFileFakerOleFakeLinkWebFakeExecetc.
  • AutoRunnerAuto-runDropFileAutoRunDownloadFileAutoRunetc.
  • Worm · BotnetWorm · botMailDosetc.
  • ScriptRunnerScript executionRemoteCmdExecHiddenPSExecetc.
OUTPUT

Verdicts ready to integrate

Per-sample results are output as structured JSON for direct integration with automation systems.
An API connects to security operations platforms such as SIEM and SOAR.

  1. 01
    JSON report

    Per-sample results are output as structured JSON for direct integration with automation systems.

  2. 02
    SIEM · SOAR integration

    An API connects to security operations platforms.

Explore other technologies

Seeing it is faster than reading about it