llmasm-ananya 0.4.0

LLM Attack Surface Mapper for AI Security Testing

██╗     ██╗     ███╗   ███╗ █████╗ ███████╗███╗   ███╗
██║     ██║     ████╗ ████║██╔══██╗██╔════╝████╗ ████║
██║     ██║     ██╔████╔██║███████║███████╗██╔████╔██║
██║     ██║     ██║╚██╔╝██║██╔══██║╚════██║██║╚██╔╝██║
███████╗███████╗██║ ╚═╝ ██║██║  ██║███████║██║ ╚═╝ ██║
╚══════╝╚══════╝╚═╝     ╚═╝╚═╝  ╚═╝╚══════╝╚═╝     ╚═╝

LLM Attack Surface Mapper

LLMASM

Discover • Intercept • Exploit • Verify

Mapping the hidden attack surface of AI-powered applications.

---

Tech Stack

---

Why LLMASM?

Traditional scanners treat AI applications as black boxes.

LLMASM treats them as attack surfaces.

• Hidden AI endpoint discovery
• Prompt injection testing
• AI proxy detection
• Tool chain mapping
• Dynamic traffic interception
• LLM-assisted verification

---

Architecture

        ┌───────────────────┐
        │   Target Website  │
        └─────────┬─────────┘
                  │
     ┌────────────┴────────────┐
     │                         │

┌─────────────┐        ┌─────────────┐
│ JS Recon    │        │ Browser     │
│ Engine      │        │ Interceptor │
└──────┬──────┘        └──────┬──────┘
       │                      │
       └──────────┬───────────┘
                  │
          ┌───────▼───────┐
          │ AI Discovery  │
          └───────┬───────┘
                  │
          ┌───────▼───────┐
          │ Active Tests  │
          └───────┬───────┘
                  │
          ┌───────▼───────┐
          │ LLM Judge     │
          └───────┬───────┘
                  │
          ┌───────▼───────┐
          │ HTML Reports  │
          └───────────────┘

---

Installation

git clone https://github.com/yourusername/llmasm

cd llmasm

pip install -r requirements.txt

---

Usage

llmasm scan https://target-ai-app.com --stealth

---

Example Output

=========================================

Target      : https://target.com
AI Score    : 92
API Score   : 85

Findings:

[HIGH] Prompt Injection Surface
[HIGH] Internal AI Proxy
[MEDIUM] Hidden API Endpoint
[LOW] Framework Fingerprint

=========================================

---

Features

• Static JavaScript Analysis
• Dynamic Network Interception
• Prompt Injection Testing
• AI Endpoint Discovery
• WAF Evasion
• LLM-as-a-Judge
• Interactive Reports
• Relationship Graphs

---

Research Areas

• Prompt Injection
• System Prompt Leakage
• Hidden AI APIs
• Tool Abuse
• RAG Exposure
• Agent Chains
• Context Leakage

---

Community & Security

LLMASM welcomes contributions, security research, and responsible disclosure.

Contributing

Interested in improving LLMASM?

• Add new payloads
• Improve detection engines
• Enhance reports
• Contribute research
• Fix bugs and documentation

Please read:

• CONTRIBUTING.md

before submitting a pull request.

---

Security Reporting

If you discover a vulnerability within LLMASM itself, please follow the responsible disclosure process described in:

• SECURITY.md

Please do not publicly disclose unpatched vulnerabilities.

---

Disclaimer

This project is intended for:

• Authorized security testing
• Security research
• Educational use
• Bug bounty programs

Unauthorized testing of systems without permission may be illegal.

---

AI applications are not black boxes.

They are attack surfaces.

⭐ Star the repository if you find it useful.