ASIPE — AI Security Interceptor & Policy Engine

Real-time endpoint DLP, prompt injection defense, and security governance platform for enterprise LLM interactions.

---

🛡️ Executive Summary

ASIPE (AI Security Interceptor & Policy Engine) is an enterprise AI security platform designed to prevent data leaks (DLP), block prompt injections/jailbreaks, and enforce compliance policies when users interact with web-based LLM applications (such as ChatGPT and Gemini).

ASIPE operates via a 3-layer security model:

1. Layer 1: Browser Extension (Client-Side Interceptor) — Hooks into DOM input elements at the capture phase (useCapture: true) to halt unsafe prompt submissions before network transmission occurs.
2. Layer 2: Central Threat & Policy Engine (Backend API) — Synchronously evaluates prompt payloads for PII, API keys/secrets, and jailbreaks, calculating a composite risk score (0–100) and returning enforcement decisions (ALLOW, REDACT, BLOCK).
3. Layer 3: Admin & Governance Console (Web Dashboard) — Provides real-time threat analytics, live security event feeds, interactive policy configuration, and compliance audit exports (SOC2, GDPR, HIPAA).

---

📐 System Architecture & Data Flow

flowchart TD
    subgraph Client ["Layer 1: Browser Extension (ChatGPT & Gemini)"]
        UserPrompt[User Submits Prompt] --> DOMCapture[DOM Event Interceptor]
        DOMCapture --> |Capture Phase| Extract[Extract Prompt Text]
    end

    subgraph Engine ["Layer 2: Threat & Policy Engine (FastAPI)"]
        Extract --> |POST /api/v1/scan-prompt| Normalization[Base64 & Homoglyph Normalizer]
        Normalization --> DLP[Regex DLP & PII Scanner]
        Normalization --> Injection[Jailbreak & Injection Classifier]
        DLP --> RiskEngine[Risk Score Calculator & Decision Matrix]
        Injection --> RiskEngine
        RiskEngine --> AuditLogger[SHA-256 Hash & Redacted Audit Logger]
        AuditLogger --> |WAL Mode| SQLite[(SQLite DB)]
    end

    subgraph Enforcement ["Client-Side UI Enforcement"]
        RiskEngine --> |Return Decision JSON| ActionHandler{Enforcement Action}
        ActionHandler --> |BLOCK| Modal[Show Warning Overlay & Stop Event]
        ActionHandler --> |REDACT| Sanitize[Replace Input with Redacted Text]
        ActionHandler --> |ALLOW| Release[Release Event to LLM Application]
    end

    subgraph Dashboard ["Layer 3: Admin Console (Next.js)"]
        SQLite --> Analytics[Real-Time Metrics & Charts]
        SQLite --> LiveFeed[Live Intercepted Feed]
        PolicyManager[Policy Rule Manager] --> |PUT /api/v1/policies| SQLite
    end

Loading

---

✨ Key Features & Technical Highlights

• ⚡ Pre-Flight DOM Capture: Intercepts Enter keypresses and send button clicks using useCapture: true and event.stopImmediatePropagation(), halting network dispatches locally before LLM API serializations.
• 🔄 SPA Resilience: Implements MutationObserver to maintain listener bindings across dynamic React and Angular DOM tree updates on ChatGPT and Gemini.
• 🧹 Text Normalization: Decodes embedded Base64 strings and cleans Unicode homoglyphs prior to executing DLP scanning logic.
• 🔑 DLP & Secret Detection: Scans for Credit Card / PAN numbers, Social Security Numbers (SSN), Email addresses, AWS Access Keys, OpenAI API Keys, and RSA Private Keys.
• 🚨 Prompt Injection Defense: Detects System Prompt Extraction attempts, DAN (Do Anything Now) jailbreaks, Developer Mode overrides, and malicious roleplay instructions.
• 🔒 Data Privacy & Honeypot Prevention: Computes SHA-256 hashes for prompt tracking and redacts sensitive PII prior to database persistence, preventing audit_logs from becoming a security honeypot.
• 📈 Governance Dashboard: Next.js portal providing real-time telemetry, action breakdowns, rule toggles, risk sliders, and downloadable CSV compliance reports.

---

📂 Project Directory Structure

PromptShield/
├── backend/                         # Layer 2: FastAPI Threat & Policy Engine
│   ├── main.py                      # FastAPI entrypoint with CORS configuration
│   ├── config.py                    # Settings, risk thresholds, API keys
│   ├── database.py                  # SQLite WAL mode connection engine
│   ├── models/
│   │   ├── db_models.py             # SQLAlchemy ORM models (AuditLog, SecurityPolicy)
│   │   └── schemas.py               # Pydantic DTOs & validation schemas
│   ├── services/
│   │   ├── dlp_scanner.py           # Regex PII & API Key scanner with text normalizer
│   │   ├── injection_classifier.py  # Jailbreak & injection detector
│   │   ├── risk_engine.py           # Risk score (0-100) calculator & decision matrix
│   │   └── audit_logger.py          # SHA-256 hashed audit log service
│   ├── routes/
│   │   ├── scan.py                  # POST /api/v1/scan-prompt endpoint
│   │   ├── policies.py              # GET & PUT /api/v1/policies endpoints
│   │   └── logs.py                  # GET /api/v1/logs & export endpoints
│   ├── tests/
│   │   └── test_scan.py             # Pytest automated unit test suite
│   └── requirements.txt             # Python dependencies
├── extension/                       # Layer 1: Manifest V3 Browser Extension
│   ├── manifest.json                # MV3 extension configuration
│   ├── background.js                # Service Worker handling storage & API fetch
│   ├── dom_interceptor.js           # Event capture & MutationObserver script
│   ├── ui_modal.js                  # In-browser warning modal UI
│   ├── content.js                   # Main content script coordinator
│   ├── styles.css                   # Warning overlay styles
│   ├── popup.html                   # Extension popup interface
│   └── popup.js                     # Popup status handler
└── dashboard/                       # Layer 3: Next.js Admin Governance Console
    ├── package.json                 # Next.js workspace & UI dependencies
    ├── next.config.js               # Next.js configuration
    └── src/
        ├── app/
        │   ├── layout.js            # Global portal layout shell
        │   ├── page.js              # Executive Overview page
        │   ├── logs/page.js         # Security audit log inspection page
        │   └── policies/page.js     # Policy manager page
        ├── components/
        │   ├── Navbar.js            # Top navigation header
        │   ├── ThreatFeed.js        # Real-time event log stream table
        │   ├── RiskCharts.js        # Analytics visualization widget
        │   ├── PolicyToggle.js      # Rule toggle & risk score slider widget
        │   └── ExportModal.js       # Compliance CSV export dialog
        └── lib/
            └── api.js               # API service client

---

🚀 Quickstart Guide

Prerequisites

• Python 3.10+
• Node.js 18+ & npm
• Google Chrome or Microsoft Edge browser

---

1. Launch Backend API Engine (Layer 2)

# Navigate to project root
cd PromptShield

# Install backend dependencies
pip install -r backend/requirements.txt

# Launch FastAPI server
python -m uvicorn backend.main:app --reload --port 8000

The API server will start at http://127.0.0.1:8000. Interactive API documentation is available at http://127.0.0.1:8000/docs.

---

2. Install Browser Extension (Layer 1)

1. Open Chrome or Edge and navigate to chrome://extensions.
2. Toggle Developer mode ON (top right corner).
3. Click Load unpacked (top left).
4. Select the directory PromptShield/extension.
5. Open ChatGPT or Gemini. You will see the 🛡️ ASIPE Active indicator in the bottom right corner.

---

3. Launch Admin Governance Dashboard (Layer 3)

# Navigate to dashboard directory
cd PromptShield/dashboard

# Install npm dependencies
npm install

# Start Next.js development server
npm run dev

Open http://localhost:3000 in your browser to access the Admin Console.

---

🧪 Running Unit Tests

Run the automated test suite covering DLP detection, jailbreak classification, risk scoring, and header authentication:

pytest backend/tests/test_scan.py

---

⚠️ Security Boundary Notice

Note

The browser extension functions as an endpoint UX agent and security nudge. Technical users can potentially bypass browser extensions via DevTools, direct API calls, or Incognito mode. For enterprise-grade unbypassable DLP enforcement, ASIPE should be paired with a network-level Forward Proxy or Cloud Access Security Broker (CASB).