AASM — AI Attack Surface Mapper

Enterprise CLI for discovering, fingerprinting, and securing AI infrastructure.

AASM is a professional Linux CLI cybersecurity platform built for penetration testers, AI security engineers, blue teams, and red teams. It discovers and assesses AI ecosystems including Local LLMs, MCP Servers, AI Agents, AI Gateways, and AI APIs.

Designed in the philosophy of Nmap, BloodHound, and Trivy — but specialized for AI infrastructure.

---

Quick Start

# Install from PyPI
pip install aasm

# Or install from source
git clone https://github.com/hashematiyat/AASM.git
cd AASM
pip install -e .

# Run a scan
aasm scan 192.168.1.0/24

# Discover AI services on localhost
aasm discover localhost

# Audit a specific service
aasm audit http://localhost:11434

# Discover MCP servers
aasm mcp 192.168.1.0/24

---

Features

Module	Capability
Discovery	Scans networks for Ollama, Open WebUI, LM Studio, LiteLLM, vLLM, HuggingFace TGI, Flowise, and more
Fingerprinting	Deep service fingerprinting — version, models, endpoints, auth, framework
MCP Scanner	JSON-RPC 2.0 MCP server discovery with tool/resource/prompt enumeration
MCP Analyzer	Security analysis of MCP permissions, dangerous tools, and trust boundaries
Agent Analyzer	Discovers AI agent frameworks and maps capabilities and risk
Assessment	Prompt injection, system prompt disclosure, jailbreak, auth bypass testing
Attack Surface Mapper	Builds asset inventory and discovers multi-hop attack paths
Risk Engine	CVSS-style scoring with MITRE ATT&CK and OWASP LLM Top 10 mapping
Reporting	HTML, JSON, and SARIF report generation
Visualization	Graphviz DOT and Mermaid attack surface diagrams

---

CLI Commands

aasm scan 192.168.1.0/24                  # Full AI infrastructure scan
aasm discover 10.0.0.0/24                 # Quick service discovery
aasm fingerprint http://localhost:11434    # Deep fingerprint one service
aasm audit http://localhost:3000          # Security audit a service
aasm mcp 192.168.1.0/24                  # Discover & audit MCP servers
aasm agents 10.0.0.0/24                  # Discover AI agents
aasm assess http://localhost:11434        # Offensive security assessment
aasm graph scan_result.json              # Generate infrastructure graph
aasm report scan_result.json             # Generate HTML/SARIF reports
aasm risk scan_result.json               # Risk scoring & executive summary
aasm platforms                            # List supported AI platforms
aasm version                              # Show version

Global Options

--config / -c      Config file path (default: aasm.yaml or ~/.config/aasm/config.yaml)
--verbose / -v     Verbose output
--debug            Debug mode

Scan Options

aasm scan 192.168.1.0/24 \
  --ports 11434,3000,8080 \
  --profile aggressive \
  --formats json,html,sarif \
  --output ./reports \
  --no-fingerprint \
  --no-mcp \
  --no-risk

---

Supported AI Platforms

Platform	Type	Default Ports
Ollama	Local LLM	11434, 11435
Open WebUI	AI Web UI	3000, 8080
LM Studio	Local LLM	1234, 1235
LiteLLM	AI Gateway	4000, 8000
vLLM	Local LLM	8000, 8080
HuggingFace TGI	Local LLM	8080, 3000
Flowise	AI Agent	3000, 3001
OpenAI-Compatible APIs	AI API	8000, 5000, …

The plugin system allows adding new platforms without modifying core code.

---

Docker

# Build
docker build -t aasm .

# Scan a network
docker run --rm -it --network host aasm scan 192.168.1.0/24

# Save reports to host
docker run --rm -it \
  --network host \
  -v $(pwd)/reports:/home/aasm/aasm_reports \
  aasm scan 192.168.1.0/24 -o /home/aasm/aasm_reports

---

Configuration

# Generate a default config
aasm config --init

# Show effective config
aasm config --show

aasm.yaml:

version: "1"

discovery:
  timeout: 5.0
  concurrency: 50
  verify_ssl: false
  ports: [11434, 3000, 1234, 4000, 8080, 8000]

assessment:
  prompt_injection: true
  prompt_leakage: true
  jailbreak: false
  max_payloads: 20

reporting:
  output_dir: ./aasm_reports
  formats: [json, html]

profiles:
  quick:
    name: quick
    ports: [11434, 3000, 1234, 4000]
    timeout: 3.0
    concurrency: 100

Use profiles: aasm scan 192.168.1.0/24 --profile aggressive

---

Plugin System

Add new platform detectors or assessment modules without touching core code:

# my_plugin.py
from aasm.plugins.base import DetectorPlugin
from aasm.core.models import AIService, AIServiceType

class MyCustomDetector(DetectorPlugin):
    name = "my-platform"
    version = "1.0.0"
    description = "Detects MyCustom AI platform"
    platform_name = "MyCustom"
    default_ports = [9999]

    async def detect(self, host: str, port: int) -> AIService | None:
        # Your detection logic here
        ...

Load plugins:

aasm plugins --load ./my_plugins/

---

Report Formats

Format	Description
JSON	Machine-readable full scan result (suitable for SIEM ingestion)
HTML	Professional dark-themed HTML report with risk matrix
SARIF	Static Analysis Results Interchange Format (GitHub, Azure DevOps)

---

Security Assessments

⚠ Only assess systems you own or have explicit written permission to test.

# Standard assessment (prompt injection + auth bypass)
aasm assess http://localhost:11434

# Full offensive assessment (includes jailbreak testing)
aasm assess http://localhost:11434 --jailbreak

# Targeted test
aasm assess http://localhost:11434 --no-prompt-injection --auth-bypass

OWASP LLM Top 10 2025 coverage:

• LLM01 — Prompt Injection
• LLM02 — Sensitive Information Disclosure
• LLM06 — Excessive Agency
• LLM07 — System Prompt Leakage
• LLM08 — Weak Guardrails

---

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v --cov=aasm

# Lint
ruff check aasm/

# Type check
mypy aasm/

---

Architecture

aasm/
├── aasm/
│   ├── cli/                    # Typer CLI commands
│   │   ├── main.py             # App entry point
│   │   ├── output.py           # Rich terminal output
│   │   └── commands/           # One file per subcommand
│   ├── core/                   # Shared domain
│   │   ├── config.py           # YAML config & profiles
│   │   ├── logger.py           # Enterprise logging
│   │   └── models.py           # Pydantic domain models
│   ├── modules/                # Feature modules (9 modules)
│   │   ├── discovery/          # Module 1 — AI Discovery Engine
│   │   │   └── platforms/      # Per-platform detectors (plugin-able)
│   │   ├── fingerprint/        # Module 2 — Fingerprinting Engine
│   │   ├── mcp/                # Module 3 — MCP Scanner & Analyzer
│   │   ├── agents/             # Module 4 — AI Agent Analyzer
│   │   ├── assessment/         # Module 5 — Security Assessment Engine
│   │   ├── mapper/             # Module 6 — Attack Surface Mapper
│   │   ├── risk/               # Module 7 — Risk Engine
│   │   ├── reporting/          # Module 8 — Reporting Engine
│   │   └── visualization/      # Module 9 — Visualization Engine
│   └── plugins/                # Plugin framework
├── tests/                      # pytest test suite
├── config/                     # Default YAML configs
├── Dockerfile
├── pyproject.toml
└── README.md

---

MITRE ATT&CK & OWASP Mapping

AASM maps all findings to:

• MITRE ATT&CK (T-codes for traditional TTPs)
• MITRE ATLAS (AML-codes for AI/ML specific threats)
• OWASP LLM Top 10 2025 for AI-specific vulnerability categories

---

Roadmap

• PDF report generation
• Neo4j graph export for BloodHound-style visualization
• MITRE ATLAS full mapping
• Kubernetes AI workload discovery
• Real-time streaming scan output
• CI/CD pipeline integration mode
• Langchain/LangGraph agent detection
• OpenTelemetry tracing
• Web dashboard (optional companion)

---

License

MIT — See LICENSE

---

AASM is intended for authorized security testing only. Always obtain written permission before scanning networks or systems you do not own.