SiegeWare Educational Lab

A declarative, reproducible cybersecurity simulation environment for studying AI-assisted offensive and defensive security practices.

┌─────────────────────────────────────────────────────────────────────────────┐
│                       Host Machine (NixOS)                                  │
│                                                                             │
│  ┌─────────────────────────────┐                                            │
│  │ Ollama Container            │   ← GPU/CPU • Runs red & blue agents     │
│  │ (localhost:11434)           │                                            │
│  └─────────────────────────────┘                                            │
│                                                                             │
│  ┌───────────────┐   ┌───────────────┐   ┌───────────────┐   ┌───────────────┐
│  │ Red Team VM   │   │ Blue Team VM  │   │ Target VM     │   │ Vulnerable VM │
│  │ 10.0.0.101    │   │ 10.0.0.102    │   │ 10.0.0.103    │   │ 10.0.0.104    │
│  │ Offensive     │   │ Defensive     │   │ Clean victim  │   │ Weak system   │
│  └───────────────┘   └───────────────┘   └───────────────┘   └───────────────┘
│         │                   │                   │                   │
│         └───────────────────┴───────────────────┴───────────────────┘
│                           br0 bridge (10.0.0.1/24)                         │
│                                                                             │
│                       DNS Controller VM (10.0.0.5)                         │
│                           • BIND9 authoritative DNS                          │
│                           • All name resolution goes through here           │
└─────────────────────────────────────────────────────────────────────────────┘

Simple I/O Flow:
You → lab-ctl CLI → Ollama (AI brain) → Red/Blue agent thinks → calls safe tools
→ runs inside MicroVM → result loops back → you see output
All network stays inside br0 — no real Internet, no breakout.

graph TD
    A[You / lab-ctl CLI] --> B[Ollama Container<br>GPU/CPU Inference]
    B --> C[Red Team Agent<br>10.0.0.101]
    B --> D[Blue Team Agent<br>10.0.0.102]
    C --> E[Target VM<br>10.0.0.103]
    C --> F[Vulnerable VM<br>10.0.0.104]
    D --> F
    E --> G[DNS Controller<br>10.0.0.5<br>BIND9]
    F --> G
    subgraph "Isolated Lab Network"
        E
        F
        G
        C
        D
    end
    subgraph "Host"
        B
        A
    end

Loading

Introducing SiegeWare

The Concept

SiegeWare (noun) /siːdʒˈwɛər/
A term coined by DeMoD LLC in 2025 to describe AI-powered autonomous cyber warfare simulation platforms designed for controlled, ethical, and educational replication of real-world offensive and defensive cybersecurity operations.

SiegeWare platforms are defined by:

• Autonomous or semi-autonomous AI agents performing red team (offensive) and blue team (defensive) roles
• Realistic, isolated network topology built with reproducible infrastructure
• Hardware-accelerated local large language model (LLM) inference for agent reasoning and decision-making
• Progressive, outcome-focused learning modules with verifiable assessment
• Strict safety boundaries that prevent any real-world harm or external impact
• Explicit emphasis on both technical mastery and ethical responsibility

Why "SiegeWare"?

The name combines two evocative elements:

• Siege — referencing the historical military tactic of surrounding and methodically reducing a fortified position, paralleling modern cyber campaigns that involve reconnaissance, persistence, lateral movement, and objective achievement.
• Ware — derived from "software" and "malware", underscoring that this is a purely software-defined, AI-augmented simulation environment, not physical warfare.

SiegeWare represents the responsible convergence of artificial intelligence, cybersecurity training, and ethical simulation technology — a digital training platform that prepares defenders and helps students understand adversary tactics without ever crossing into real-world harm.

This Lab as a SiegeWare Simulator

The Universal Educational AI Agents Lab is intentionally engineered as a full-featured SiegeWare simulator, with the core mission of preparing the next generation of IT security professionals.

Key SiegeWare characteristics implemented in this platform:

1. Autonomous Agent Behavior
   Red and blue team agents leverage local LLMs to independently reason, plan, and execute actions within the simulation.

2. Realistic Attack/Defense Lifecycle
   Five progressive labs mirror actual cyber kill chains, defensive workflows, and incident response processes.

3. Isolated, High-Fidelity Environment
   MicroVMs, containerized services, and virtual networking create production-like conditions while guaranteeing complete isolation.

4. Hardware-Agnostic Scalability
   Runs natively on consumer laptops (Apple Silicon via Asahi Linux, Intel/AMD/NVIDIA GPUs) and scales to classroom servers or research clusters.

5. Verifiable Educational Outcomes
   Structured objectives, automated verification scripts, progress tracking, and instructor oversight tools.

6. Ethical & Safety Framework
   Explicit system prompts, network containment, no external connectivity, and repeated emphasis on responsible use and simulation-only actions.

Target Audience & Educational Impact

This SiegeWare simulator is designed for:

• University cybersecurity programs (undergraduate and graduate levels)
• Professional training organizations (SANS, Offensive Security, EC-Council, etc.)
• Corporate red team / blue team / purple team training programs
• Independent learners preparing for certifications (OSCP, PNPT, CRTP, CEH, etc.)
• AI security researchers studying agent behavior in adversarial environments

By releasing this platform as free, open-source software under GPL-3.0, DeMoD LLC seeks to:

• Democratize access to high-fidelity AI-augmented cybersecurity training
• Accelerate the development of AI-literate security professionals
• Promote ethical understanding of both offensive and defensive capabilities
• Establish a widely adopted reference platform for modern AI-security education

SiegeWare is not entertainment. It is a professional training environment.

The future of cybersecurity will be shaped by those who understand both how AI can attack systems and how AI can defend them. This lab exists to train that next generation — safely, responsibly, and effectively.

Overview

This lab offers a secure, fully isolated environment to study the application of artificial intelligence in cybersecurity through structured red team vs. blue team simulations. Built with Nix and NixOS, it ensures complete reproducibility across deployments and supports a wide range of hardware platforms.

Learning Objectives

By completing the full lab series, participants will be able to:

1. Use AI-assisted methods to perform network reconnaissance and asset enumeration
2. Identify and safely simulate exploitation of privilege escalation vectors
3. Implement defensive monitoring, anomaly detection, and incident response using AI agents
4. Execute multi-stage attack campaigns while practicing operational security
5. Optimize autonomous AI agents for competitive red vs. blue scenarios
6. Critically evaluate the role, strengths, limitations, and ethical implications of AI in security operations

Core Capabilities

• Fully declarative infrastructure — 100% reproducible via Nix flakes
• Broad hardware compatibility — x86_64 (NVIDIA CUDA, AMD ROCm, Intel Arc, CPU) and aarch64 (Apple Silicon Metal)
• GPU-accelerated inference — Local LLMs via Ollama
• Strong isolation — MicroVM-based execution environments with network containment
• Structured curriculum — Five progressive labs with clear objectives and verification
• Integrated management tooling — lab-ctl CLI for students and instructors

Deployment Instructions

Prerequisites

• Nix package manager (https://nixos.org/download)
• Hardware:
  • Minimum: 16 GB RAM, 4-core CPU
  • Recommended: 32+ GB RAM, GPU (NVIDIA/AMD/Intel) or Apple Silicon
• Basic Linux terminal proficiency

Step-by-Step Deployment

Best plug and play UX with Oligarchy NixOS lite

1. Clone the repository

   git clone https://github.com/ALH477/ai-agents-lab.git
   cd ai-agents-lab

2. Deploy the lab infrastructure

   nix run .#deploy

   This command:

   • Builds and activates the NixOS configuration
   • Starts the Ollama inference container
   • Launches MicroVMs (red-team, blue-team, target, vulnerable-vm, dns-controller)
   • Configures isolated networking and DNS authority

3. Verify deployment

   nix run .#status

   Expected output includes:

   • Ollama service running
   • All MicroVMs active
   • DNS controller responding (dig @10.0.0.5 red.lab.local)
   • Loaded models listed

4. Access student guide

   nix run .#student-quickstart

5. (Optional) Build portable Docker image

   nix build .#inferenceImage
   docker load < result

Post-Deployment Checks

• Ollama API: curl http://localhost:11434/api/tags
• DNS resolution: dig @10.0.0.5 red.lab.local
• VM connectivity: ping 10.0.0.101 (from host or another VM)

Laboratory Exercises

Lab 01: Network Reconnaissance

Level: Foundational | Duration: 30–45 minutes | Points: 100

Focus: AI-assisted enumeration and intelligence gathering
Key Skills: Port scanning, service fingerprinting, OS detection, banner grabbing
Learning Outcomes: Understand reconnaissance phase of penetration testing; interpret scan results; apply AI for tool selection and analysis

Lab 02: Privilege Escalation

Level: Intermediate | Duration: 60–90 minutes | Points: 150

Focus: Identification and simulation of privilege escalation vectors
Key Skills: SUID/SGID binary analysis, permission misconfiguration, service exploitation
Learning Outcomes: Recognize common escalation paths; assess risk of misconfigurations; practice controlled exploitation

Lab 03: Security Monitoring & Detection

Level: Intermediate | Duration: 45–60 minutes | Points: 125

Focus: Defensive operations and anomaly detection
Key Skills: Log analysis, network monitoring, alert rule creation
Learning Outcomes: Build foundational detection capabilities; understand blue team workflows; apply AI to accelerate threat identification

Lab 04: Advanced Red Team Operations

Level: Advanced | Duration: 90–120 minutes | Points: 200

Focus: Execution of multi-stage attack campaigns
Key Skills: Stealth reconnaissance, persistence, lateral movement, data exfiltration
Learning Outcomes: Conduct structured attacks; apply operational security practices; understand evasion techniques

Lab 05: Autonomous AI Red vs Blue Competition

Level: Advanced | Duration: 120+ minutes | Points: 300

Focus: Strategy optimization for competing autonomous AI agents
Key Skills: Prompt engineering, performance tuning, attack-defense balance
Learning Outcomes: Explore emergent behavior in AI security systems; understand trade-offs between aggression and stealth

Instructor Guide

Role & Responsibilities

Instructors serve as facilitators of learning, not just content deliverers. Your role includes:

• Setting clear expectations for ethical use
• Monitoring student progress and intervening when needed
• Providing context and real-world relevance
• Assessing learning outcomes fairly and consistently
• Customizing labs to match course objectives

Preparation Checklist

1. Deploy the lab on instructor workstation/server

   nix run .#deploy

2. Verify all components

   nix run .#status

3. Run instructor setup script

   nix run .#instructor-setup

4. Create student accounts or environments (future feature)

   • Current: single shared lab (recommended for initial classes)
   • Planned: per-student MicroVM cloning

5. Review lab materials

   • /var/lib/ai-agents-lab/labs/
   • Each lab has lab.json, starter.py, verify.py

Class Session Structure (Recommended)

Duration: 2–3 hours per lab (including debrief)

1. Introduction (10–15 min)

   • State learning objectives
   • Review ethical guidelines
   • Explain lab controller commands

2. Guided Start (15–20 min)

   • Students run lab-ctl student start <lab-id>
   • Instructor walks through starter code

3. Independent Work (60–120 min)

   • Students interact with agents
   • Use lab-ctl student status and verify
   • Instructor circulates, answers questions

4. Debrief & Discussion (20–30 min)

   • Share findings (screenshots, agent conversations)
   • Discuss what worked, what failed
   • Highlight defensive lessons from offensive actions

5. Assessment

   • Run lab-ctl student verify on student machines
   • Instructor reviews outputs
   • Use lab-ctl instructor grade <student-id> (future)

Monitoring & Intervention

• Real-time monitoring:

  lab-ctl instructor monitor student-01

• Class-wide stats:

  lab-ctl instructor stats

• Reset stuck student:

  lab-ctl instructor reset student-01

Customization Tips

• Add new labs by creating directories under packages/lab-controller/labs/
• Modify objectives/hints in lab.json
• Extend verification logic in verify.py
• Adjust DNS records in modules/ai-agents-env.nix for custom domains

Grading Recommendations

• 40% Objective completion (via verify)
• 30% Quality of documentation / notes
• 20% Ethical reasoning demonstrated
• 10% Creativity / sophistication of approach

System Architecture

┌─────────────────────────────────────────────────────────────┐
│                     Host System                              │
│  ┌────────────────────────────────────────────────────────┐ │
│  │  Ollama Container (GPU-accelerated inference)          │ │
│  │  - red-qwen-agent   (Offensive AI)                     │ │
│  │  - blue-llama-agent (Defensive AI)                     │ │
│  └────────────────────────────────────────────────────────┘ │
│                                                              │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌───────────────┐
│  │  Red Team   │  │  Blue Team  │  │   Target    │  │  Vulnerable │  │ DNS Controller│
│  │  MicroVM    │  │  MicroVM    │  │   MicroVM   │  │     VM      │  │   (BIND9)     │
│  │  10.0.0.101 │  │  10.0.0.102 │  │  10.0.0.103 │  │  10.0.0.104 │  │   10.0.0.5    │
│  └─────────────┘  └─────────────┘  └─────────────┘  └─────────────┘  └───────────────┘
│         │                 │                 │                 │                 │
│         └─────────────────┴─────────────────┴─────────────────┴─────────────────┘
│                     br0 (10.0.0.1/24) – Isolated Lab Network                   │
└─────────────────────────────────────────────────────────────┘

Security & Ethical Considerations

Isolation & Safety Features

• MicroVMs provide kernel-level isolation from host
• Network traffic confined to virtual bridge (br0)
• No direct Internet access from any VM
• DNS resolution controlled by isolated DNS controller
• Environment fully resettable via rebuild
• All configurations declarative and auditable

Responsible Use Policy

Participants must:

• Use knowledge gained solely for authorized educational or professional purposes
• Never apply techniques to production systems without explicit permission
• Adhere strictly to ethical and legal guidelines
• Report discovered vulnerabilities responsibly

Recommended Learning Path

Module 1: Foundations (Weeks 1–2)

• Lab 01 – Network Reconnaissance
• Focus: AI-assisted enumeration, tool selection

Module 2: Offensive Security (Weeks 3–4)

• Lab 02 – Privilege Escalation
• Focus: Vulnerability identification, exploitation techniques

Module 3: Defensive Security (Weeks 5–6)

• Lab 03 – Security Monitoring & Detection
• Focus: Log analysis, anomaly detection, alerting

Module 4: Advanced Operations (Weeks 7–8)

• Lab 04 – Multi-stage Attack Campaigns
• Focus: Persistence, lateral movement, evasion

Module 5: Autonomous Systems (Weeks 9–10)

• Lab 05 – AI Red vs Blue Competition
• Focus: Strategy optimization, emergent behavior

Educational Outcomes

By completing this lab series, participants will develop:

Technical Competencies

• Network reconnaissance and enumeration
• Privilege escalation analysis
• Security monitoring and incident detection
• Multi-stage attack execution
• AI agent orchestration and tuning

Conceptual Understanding

• Attack surface mapping
• Defense-in-depth principles
• Operational security (OpSec)
• AI limitations in security contexts
• Ethical considerations in offensive security

Professional Skills

• Structured documentation and reporting
• Risk assessment and prioritization
• Responsible disclosure practices
• Prompt engineering for security tasks

Troubleshooting Guide

Ollama Not Responding

sudo systemctl status docker-inference-optimized
sudo journalctl -u docker-inference-optimized -f
sudo systemctl restart docker-inference-optimized

MicroVMs Not Starting

systemctl list-units 'microvm@*' --no-pager
sudo systemctl start microvm@red-team
sudo systemctl start microvm@blue-team
sudo systemctl start microvm@target
sudo systemctl start microvm@vulnerable-vm
sudo systemctl start microvm@dns-controller

DNS Resolution Issues

sudo systemctl status microvm@dns-controller
dig @10.0.0.5 red.lab.local
ssh root@10.0.0.101 "dig red.lab.local"

Lab Controller Issues

which lab-ctl
ls -la /var/lib/ai-agents-lab/labs/
sudo chmod -R 755 /var/lib/ai-agents-lab/

Network Connectivity

ip addr show br0
sudo systemctl status dhcpd4
sudo systemctl restart systemd-networkd

Contributing

Adding New Labs

1. Create lab directory structure
2. Define lab.json metadata
3. Add optional starter.py, verify.py, README.md
4. Test with:

   lab-ctl student start <new-lab-id>
   lab-ctl student verify

Reporting Issues

Provide:

• Clear problem description
• Steps to reproduce
• Expected vs. observed behavior
• System info (architecture, Nix version, hardware)

License

Copyright © 2025 DeMoD LLC
Licensed under the GNU General Public License v3.0 (GPL-3.0)

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/.

Acknowledgments

• Ollama project — efficient local LLM inference
• MicroVM.nix — lightweight virtualization
• NixOS ecosystem — declarative system configuration
• Educational contributors and reviewers

Contact & Support

• Documentation — This README
• Troubleshooting — See section above
• Issues — GitHub issue tracker
• Community — (Discord/forum link forthcoming)

This SiegeWare simulator is designed to support rigorous, structured learning in AI security. Feedback and contributions are welcome to enhance its educational impact.