SPECTER CRYPT

AI-assisted ransomware simulation — authorised adversarial testing. Recon. Encrypt. Extort. Propagate. Eight subsystems.
8
Subsystems
AES-256
Encryption
2
C2 Channels
297
Tests Passing
pip install red-specter-specter-crypt
Documentation
AI writes the ransom note / LLM APIs route C2 traffic / Shadow copies deleted before encryption starts / Double extortion: exfil then encrypt / Pass-the-hash lateral movement in minutes / Key escrow enables selective decryption / Whitespace steganography hides beacon traffic / Ransomware simulation requires real encryption logic AI writes the ransom note / LLM APIs route C2 traffic / Shadow copies deleted before encryption starts / Double extortion: exfil then encrypt / Pass-the-hash lateral movement in minutes / Key escrow enables selective decryption / Whitespace steganography hides beacon traffic / Ransomware simulation requires real encryption logic

Modern Ransomware Is AI-Enhanced

Modern ransomware campaigns combine AI-generated social engineering, covert LLM-API command-and-control channels, double-extortion data staging, and automated lateral movement. Defenders building detection and response capabilities need to test against realistic simulations — not theoretical models. SPECTER CRYPT operationalises the full ransomware kill chain in a controlled, authorised red-team environment.

LLM APIs as Covert C2

Ransomware operators route command-and-control traffic through legitimate LLM API calls — base64-encoded JSON payloads embedded in model prompts, responses carrying instructions. Whitespace steganography hides signals in normal text. API traffic blends with legitimate AI usage: no firewall blocks it.

Shadow Copy Destruction

Before any file is encrypted, production ransomware deletes Volume Shadow Copies via vssadmin and wmic, stops backup services, and removes restore points. Defenders who have not tested VSS destruction detection have a blind spot that ransomware actors exploit in every major campaign.

Double Extortion

Exfiltrate high-value files before encrypting them. Victims face two pressure points: pay to decrypt, and pay to suppress publication. SPECTER CRYPT stages sensitive files, creates encrypted archives, and simulates the exfil channel — giving defenders realistic telemetry to tune detection rules against.

AI-Generated Ransom Notes

LLM-generated ransom notes are grammatically perfect, contextually tailored, and psychologically optimised. They reference specific files, quote revenue figures, and include negotiation portals. AI-written notes are harder for victims to dismiss — and require different detection and response playbooks.

Pass-the-Hash Propagation

Ransomware spreads laterally using harvested NTLM hashes — no plaintext password required. Impacket psexec and pass-the-hash techniques pivot from an initial foothold to every reachable host in minutes. Defenders need real propagation simulations to validate network segmentation and lateral movement detection.

No Simulation Evidence for IR Exercises

Incident response drills without realistic artefacts — real encrypted files, real C2 traffic, real ransom notes — produce unrealistic outcomes. SPECTER CRYPT generates complete, signed evidence chains for IR team training, red-team exercises, and authorised penetration testing engagements.

The SPECTER CRYPT Engine

Eight subsystems cover the complete ransomware kill chain — from initial reconnaissance through encryption, extortion, and lateral propagation. DESTROY-tier gates enforce authorisation requirements for the most destructive operations.

# Subsystem Command Gate What It Does
01 RECON specter-crypt recon OPEN Target environment enumeration. File system analysis, high-value file identification (office docs, databases, configs, credentials), network share discovery, connected host enumeration. Outputs target manifest for downstream subsystems.
02 SHADOW specter-crypt shadow INJECT Backup and shadow copy destruction simulation. VSS deletion via vssadmin/wmic command generation, Windows backup service stop sequence, restore point removal. Dry-run mode generates command artefacts without executing; live mode requires explicit --confirm-inject flag.
03 EXFIL specter-crypt exfil INJECT Pre-encryption data staging for double-extortion simulation. High-value file selection from RECON manifest, AES-256-CBC archive creation, staging directory management. Simulates exfiltration channel with configurable destination. Generates double-extortion evidence artefacts.
04 C2 specter-crypt c2 INJECT LLM-API covert command-and-control channel. Two modes: base64_json (operator instructions encoded as JSON, embedded in LLM prompt payloads, decoded from model responses) and whitespace_stego (beacon signal encoded as Unicode zero-width characters in cover text). Supports OpenAI, Anthropic, Gemini. PCAP-compatible traffic generation.
05 RANSOM specter-crypt ransom INJECT AI-generated ransom note production. LLM-authored notes in corporate, personal, and government templates — contextually tailored with target organisation name, file counts, and urgency framing. Negotiation bot portal URL generation (onion address format). Bitcoin/Monero payment address templating.
06 ENCRYPT specter-crypt encrypt DESTROY AES-256-CBC file encryption with key escrow. Per-file IV generation, PKCS7 padding, .crypt extension appending. Master key stored in encrypted escrow file (RSA-wrapped or passphrase-derived). Selective targeting by file type, directory, or RECON manifest. Companion decrypt command for authorised recovery.
07 PROPAGATE specter-crypt propagate DESTROY Lateral movement simulation via impacket. Two methods: psexec (service-based remote execution on SMB targets) and pass_the_hash (NTLM hash-based authentication without plaintext credentials). Target host list from RECON manifest. Deploys simulation payload to reachable hosts under authorised test conditions.
08 REPORT specter-crypt report OPEN Ed25519-signed CryptReport (SCR-{hex12}). SHA-256 hash-chained evidence across all subsystem artefacts. MITRE ATT&CK TTP mapping. OWASP LLM taxonomy. Severity ratings per finding. JSON output with full artefact manifest for IR team handoff.

One Command. Complete Ransomware Simulation.

Run the complete pipeline — recon to signed report — against an authorised target environment:

$ specter-crypt run --target ./testenv/ --c2-provider openai --c2-mode base64_json --override --confirm-destroy
[RECON] Enumerating target environment...
  Files: 4,237 | High-value: 847 | Shares: 3 | Network hosts: 12
  File types: docx/xlsx/pdf/sql/env/pem | Manifest: recon_manifest.json
[SHADOW] Generating shadow copy destruction commands...
  VSS: 4 copies targeted | WMIC: generated | Backup svc: stop sequence ready
[EXFIL] Staging high-value files for double extortion...
  Files staged: 847 | Archive: exfil_stage.tar.gz.enc | Size: 2.3 GB
[C2] Establishing LLM-API covert C2 (base64_json)...
  Provider: openai | Payload: base64_json embedded | Beacon: ACKNOWLEDGED
  Whitespace stego test: 128-token cover | Signal ratio: 0.94 | PCAP: saved
[RANSOM] Generating AI ransom note (corporate template)...
  LLM: contextually tailored | Portal: scr-a3f29b1c.onion | Note: ransom.txt
[ENCRYPT] Encrypting target files (AES-256-CBC)...
  Files encrypted: 847 | Extension: .crypt | IV: per-file
  Key escrow: saved to escrow.key.enc | Decrypt ready: specter-crypt decrypt
[PROPAGATE] Lateral movement via impacket psexec...
  Host: 10.0.0.15 | Method: psexec | Status: shell acquired
  Pass-the-hash: 10.0.0.23 | NTLM: accepted | Propagated: 2/12 hosts

COMPLETE | Report: SCR-a3f29b1c4e7d | Signed ✓ | Severity: CRITICAL

AES-256-CBC + Key Escrow

Real AES-256-CBC encryption — per-file IVs, PKCS7 padding, .crypt extensions. Master key in encrypted escrow. Companion decrypt command for authorised recovery after red-team exercises.

LLM-API Covert C2

Two channels: base64_json encodes operator instructions in LLM prompts — decoded from model responses. Whitespace steganography hides beacon signals in Unicode zero-width characters inside normal cover text.

Ed25519 Signed Reports

Every CryptReport cryptographically signed with Ed25519. SHA-256 hash-chained evidence across all artefacts. Tamper-evident by design. Unique report ID: SCR-{hex12}. IR team handoff ready.

Three UNLEASHED Tiers

OPEN (recon/report), INJECT (shadow/exfil/c2/ransom — --override), DESTROY (encrypt/propagate — --override --confirm-destroy). Ed25519 cryptographic gate enforced at each tier boundary.

8
Subsystems
AES-256
Cipher
2
C2 Channels
297
Tests Passing
0
Failures

LLM APIs as Command-and-Control Infrastructure

SPECTER CRYPT implements two covert C2 channels using commercial LLM APIs as the transport layer. Both channels blend operator traffic with legitimate AI API usage — making them invisible to conventional network monitoring and firewall rules that trust LLM API endpoints.

base64_json — Prompt Embedding

  • Operator instructions serialised as JSON
  • Base64-encoded, embedded in LLM prompt text
  • Model response carries encoded acknowledgement
  • Client decodes response to extract C2 commands
  • Traffic indistinguishable from normal API calls
  • Supports: OpenAI · Anthropic · Gemini
  • PCAP capture included for IR training use

whitespace_stego — Unicode Steganography

  • Beacon signal encoded as zero-width characters
  • Injected into normal cover text (128–256 tokens)
  • Receiver decodes hidden signal from whitespace pattern
  • Cover text appears entirely normal to human readers
  • Signal ratio measurement: encoder fidelity scoring
  • Detectable only with Unicode normalisation inspection
  • Tests defensive whitespace-stego detection rules
OpenAI
gpt-4o-mini (default)
gpt-4o · gpt-4-turbo
Anthropic
claude-3-5-haiku
claude-3-5-sonnet
Gemini
gemini-1.5-flash
gemini-1.5-pro
Generic
Ollama · vLLM
Any OpenAI-compat

Real AES-256-CBC. Real Key Escrow.

SPECTER CRYPT implements production-grade AES-256-CBC encryption — the same cipher used by real ransomware families. Key escrow enables authorised recovery after red-team exercises. The decrypt command is always available to operators holding the escrow key.

Encryption — AES-256-CBC

  • AES-256-CBC with per-file IV generation
  • PKCS7 padding, .crypt extension on encrypted files
  • Selective targeting: file type / directory / manifest
  • Master encryption key generated per-campaign
  • Original files overwritten with encrypted versions
  • Ransom note dropped in each affected directory
  • Requires: --override --confirm-destroy (DESTROY gate)

Key Escrow — Authorised Recovery

  • Master key stored in escrow.key.enc
  • RSA-wrapped or passphrase-derived escrow format
  • specter-crypt decrypt restores all affected files
  • Per-file IV stored in encrypted file header
  • Recovery verified against SHA-256 pre-encryption hashes
  • Audit log of every encrypted/decrypted file path
  • No files permanently lost in authorised exercises

Propagation — impacket

  • psexec: service-based remote execution over SMB
  • pass_the_hash: NTLM hash auth, no plaintext needed
  • Target list from RECON network host manifest
  • Simulation payload deployed to each reachable host
  • Per-host propagation result and timing recorded
  • Requires: --override --confirm-destroy (DESTROY gate)
  • Scope file enforces authorised host list

Every Finding Mapped

MITRE ATT&CK

Adversarial Tactics

  • T1486 — Data Encrypted for Impact (ENCRYPT subsystem)
  • T1490 — Inhibit System Recovery (SHADOW subsystem)
  • T1041 — Exfiltration Over C2 Channel (EXFIL + C2)
  • T1570 — Lateral Tool Transfer (PROPAGATE psexec)
  • T1550.002 — Pass the Hash (PROPAGATE pth)
  • T1566 — Phishing / Social Engineering (RANSOM)
MITRE ATLAS

AI-Specific Tactics

  • AML.T0051 — LLM Prompt Injection (C2 channel)
  • AML.T0043 — Craft Adversarial Data (ransom note gen)
  • AML.T0056 — LLM Jailbreak (C2 evasion techniques)
  • AML.T0048 — External Harms (double extortion impact)
  • AML.T0040 — Supply Chain Compromise (propagation)
OWASP LLM

LLM Security Taxonomy

  • LLM01 — Prompt Injection (C2 base64_json channel)
  • LLM02 — Insecure Output Handling (ransom note gen)
  • LLM06 — Excessive Agency (autonomous propagation)
  • LLM08 — Vector & Embedding Weaknesses (stego C2)
  • LLM09 — Misinformation (AI-generated extortion content)

Authorised Use Only

SPECTER CRYPT is designed exclusively for authorised red-team exercises, penetration testing engagements, and ransomware defence validation in controlled environments. Use against production systems, unauthorised hosts, or without explicit written permission from the system owner is illegal and strictly prohibited. The encryption subsystem generates real encrypted files — always run in a dedicated test environment with full backups. Unauthorised use may violate Computer Misuse Act 1990 (UK), Computer Fraud and Abuse Act (US), and equivalent legislation globally. SPECTER CRYPT is restricted to licensed operators with verified UNLEASHED credentials. Apache License 2.0.

Pure Engineering
Real Encryption. Real C2. Real Propagation.

SPECTER CRYPT performs live AES-256-CBC encryption, makes real LLM API calls for C2 channels, and executes real impacket lateral movement against authorised targets. Every subsystem connects to real infrastructure. UNLEASHED fires real payloads. Tests passing is not proof — live kill chain execution is.

8
Kill Chain Subsystems
256
AES Key Bits
2
Covert C2 Channels
297
Tests Passing
Ed25519 Cryptographic Gate
SPECTER CRYPT UNLEASHED

Three tiers. Ed25519 key required for INJECT and DESTROY. Real encryption and lateral movement gated at the highest tier. One operator. One key.

OPEN
RECON · REPORT
No flags required
INJECT
SHADOW · EXFIL · C2 · RANSOM
--override required
DESTROY
ENCRYPT · PROPAGATE
--override --confirm-destroy