AI Security Posture Management

01Overview

Traditional SAST tools were built for web apps and miss the entire class of risks that AI/ML systems introduce. AI Security Posture Management (AI-SPM) closes that gap: a single-file, zero-dependency Python scanner that statically analyzes AI/ML codebases end-to-end — from data ingestion and model training through deployment and live inference — and flags the security misconfigurations, unsafe patterns, and compliance gaps that generic scanners walk straight past.

Point it at a repo and it does three things at once. It runs 145 AI-specific SAST rules across 26 categories (unsafe pickle/torch.load model loading, prompt injection, data poisoning, PII leakage to LLM APIs, missing guardrails, over-autonomous agents, RAG and MCP exposure, shadow AI, and more). It builds an AI/ML inventory — auto-discovering the frameworks, models, APIs, vector databases, and experiment trackers actually present in the code. And it checks the supply chain, matching pinned dependency versions against a curated database of 33 real AI/ML CVEs in packages like TensorFlow, PyTorch, LangChain, mlflow, Ray (ShadowRay), Ollama, and ChromaDB.

Every finding is mapped to four AI-specific governance frameworks — NIST AI RMF, the EU AI Act, OWASP ML Top 10, and MITRE ATLAS — so a security finding doubles as a compliance signal. It scans Python, JS/TS, Jupyter notebooks, .env, YAML/TOML, Terraform, Dockerfiles, and dependency manifests, and emits colored console output, machine-parseable JSON, and an interactive HTML report.

It is built for AppSec engineers, ML platform teams, and AI governance leads who need to shift AI risk left. Because the whole scanner is one portable Python 3.10+ stdlib file with no pip installs, it drops into any CI pipeline (GitHub Actions, GitLab, Azure DevOps, Jenkins) and gates builds via an exit code of 1 on CRITICAL or HIGH findings — no agents, no cloud, no dependency management.

02Key Capabilities

03Architecture

The entire scanner is one self-contained Python file (ai_spm_scanner.py, ~2,230 LOC, stdlib only). Rules live as module-level lists of dicts; an AISPMScanner class walks the target, dispatches each file by extension to a language-specific scanner, runs the regex SAST engine line-by-line, and simultaneously builds an AI/ML inventory and checks dependency manifests against a CVE database. Findings (a Finding dataclass carrying CWE, CVE, and compliance tags) are then rendered to console, JSON, or HTML.

04Project Structure

05Security Controls

06Technology Stack

Language

Python 3.10+ (modern type-hint syntax with `from __future__ import annotations`; stdlib only)

Dependencies

None — zero external packages, pure standard library (re, argparse, json, dataclasses, pathlib)

Distribution

Single self-contained file (ai_spm_scanner.py, ~2,230 LOC)

Analysis Engine

Regex-based line-by-line SAST across module-level rule dicts

Output Formats

ANSI console, JSON, self-contained interactive HTML (dark purple-gradient theme)

Testing

Intentionally vulnerable fixture files under tests/samples/ for rule validation

CI/CD

Exit-code gating with documented GitHub Actions, GitLab CI, Azure DevOps, and Jenkins recipes

Platforms

Windows, macOS, Linux (no compilation, no venv required)

07Quick Start

$ git clone https://github.com/Krishcalin/AI-Secure-Posture-Management.git && cd AI-Secure-Posture-Management
$ python ai_spm_scanner.py --version   # verify: AI-SPM Scanner v1.1.0
$ python ai_spm_scanner.py /path/to/your/ai-project   # scan a whole project
$ python ai_spm_scanner.py ./my-project --json report.json --html report.html   # JSON + interactive HTML reports
$ python ai_spm_scanner.py ./my-project --severity HIGH   # only CRITICAL and HIGH findings (CI gate)
$ python ai_spm_scanner.py tests/samples/ --verbose   # run against bundled vulnerable fixtures

08Compliance & Frameworks

09Integrations & Outputs