community-ai-audit 0.1.0

Community-driven AI security audit tool using interpretability techniques

Community AI Security Audit Tool

A community-driven tool for auditing AI models for security vulnerabilities using interpretability techniques. Built for cybersecurity researchers, ML engineers, and AI safety enthusiasts.

Features

• Vulnerability Scanning — Detect backdoors, adversarial vulnerabilities, data poisoning, model stealing risks, and membership inference.
• Interpretability Analysis — Apply integrated gradients, LIME, activation clustering, and more to explain model behavior.
• Risk Scoring & Reporting — Generate actionable reports (Markdown, JSON, HTML, SARIF) with severity scores and mitigation suggestions.
• Plug & Play — Swap any model provider, SIEM, or scanner without touching the core.
• Community Sharing — Share audit results, signatures, and detection rules via a lightweight database or git-based sharing.

Installation

git clone https://github.com/your-org/community-ai-audit.git
cd community-ai-audit
pip install -e .

Quick Start

# Discover all components
community-ai-audit discover

# Scan a local PyTorch model
community-ai-audit scan my_model.pt --provider local --scanners adversarial --probe-file examples/data/toy_probe.json

# Full audit with interpretability
community-ai-audit audit my_model.pt --provider local --profile standard --scanners adversarial backdoor --interpreters integrated-gradients --probe-file examples/data/toy_probe.json --input '[0.1,...'

# Generate report in HTML
community-ai-audit audit my_model.pt --provider local --output html --save report.html

Supported Providers

Adapter	Type	Works Offline	Setup
local	PyTorch, TF, ONNX	✅ Yes	Install torch/tensorflow
huggingface	transformers, diffusers	✅ Yes	Install transformers
openai	GPT-4o, o1, etc.	❌ No	Set OPENAI_API_KEY
anthropic	Claude-3, etc.	❌ No	Set ANTHROPIC_API_KEY
aws_bedrock	All Bedrock models	❌ No	Set AWS creds
ollama	Local LLMs	✅ Yes	Install ollama

Supported SIEMs

Connector	Platform	Auth
splunk	Splunk HEC	Token
elastic	Elastic Security	API Key
datadog	Datadog Logs	API Key
sentinel	Microsoft Sentinel	Shared Key

Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Model     │────→│  Scanner(s)  │────→│  Findings   │
│   Adapter   │     └──────────────┘     └─────────────┘
│             │
│             │────→│ Interpreter  │────→│Attributions │
│             │     └──────────────┘     └─────────────┘
└─────────────┘              │                   │
                              ↓                   ↓
                         ┌────────────┐
                         │  Reporter  │
                         │ (md/json/  │
                         │  html/sf)  │
                         └────────────┘

[CLI] → [AuditEngine] → [Plugins] → [SIEM Connector]

See docs/ARCHITECTURE.md for the full architecture.

Documentation

Guide	Purpose
ADAPTER_GUIDE.md	Add a new model provider (30 min)
SCANNER_GUIDE.md	Add a new vulnerability scanner (30 min)
CONNECTOR_GUIDE.md	Add a new SIEM connector (30 min)
INTERPRETER_GUIDE.md	Add a new interpreter (30 min)
PLUGIN_GUIDE.md	Quick reference for all plugins
ARCHITECTURE.md	System design and data flow
CONTRIBUTING.md	Setup, workflow, PR checklist
PHASE1_BENCHMARK.md	Benchmark reproducibility
PHASE2_CONNECTORS.md	SIEM integration guide

Project Status

🚧 Pre-release (v0.1.0) — Core infrastructure in progress.

Contributing

See CONTRIBUTING.md for setup instructions and guidelines.

License

MIT License — see LICENSE.