sentinel-ai-scanner 0.1.4

AI model security scanner — scan any HuggingFace model for vulnerabilities, CVEs, misconfigurations, prompt injection, and bias.

SentinelAI CLI

The VirusTotal for AI Models — scan any HuggingFace model for security vulnerabilities in one command.

Features

• 🛡️ Serialization Scanner — detects hidden code execution in model files
• 📦 CVE Scanner — cross-references dependencies against the CVE database
• ⚙️ Config Auditor — flags dangerous settings like trust_remote_code
• 💉 Behavioral Probe — tests for prompt injection and jailbreak vulnerabilities
• ⚖️ Bias Check — detects demographic bias in model outputs
• 📊 Risk Score — unified 0–100 score mapped to OWASP LLM Top 10

Install

pip install sentinel-ai-scanner

Quick Start

# 1. Log in via GitHub OAuth
sentinelai login

# 2. Scan any HuggingFace model
sentinelai scan meta-llama/Llama-3-8B

# 3. Log out when done
sentinelai logout

Commands

Command	Description
sentinelai login	Authenticate via GitHub OAuth
sentinelai scan <model>	Scan a model (HuggingFace ID or local path)
sentinelai logout	Remove stored credentials

Scan Options

sentinelai scan meta-llama/Llama-3-8B \
  --hf-token hf_xxx... \      # For private models
  --output-dir ./reports       # Save reports here

Output

After scanning, you get:

• Terminal output with live progress and risk score
• JSON report for CI/CD pipelines
• HTML report for human-readable sharing
• Dashboard link to view results on the web

Requirements

• Python 3.10+
• Works on macOS, Linux, and Windows

Links

• 🌐 Web Dashboard
• 📖 GitHub Repository
• 🐛 Report Issues

License

MIT