Skip to content

Repository files navigation

Shadow AI Audit Kit

Find out which AI tools are actually being used on a machine — in minutes, from a USB stick.

Your staff are using AI tools you don't know about. Client files pasted into chatbots. Customer data in prompts. A browser history full of services your policies never approved. This kit answers the question every business owner should be asking: what is actually happening on our machines?

Plug in the USB stick (or clone this repo), run one script, and get a risk-ranked report of AI-tool usage across browsers, installed software, and DNS logs. No cloud dependency. No telemetry. Nothing leaves the machine.

Built and field-tested by elect-rix Technology Solutions for professional Shadow AI Discovery audits. Released open-source so you can verify exactly what it does — and run it yourself.

This GitHub tree is the public source. It is not a second product and not a substitute for an elect-rix field USB stick. Field sticks carry hashed platform binaries (bin/linux, bin/mac, bin/windows) built from this code. If a checkout and a stick disagree, trust the hashed stick — or rebuild from scanner.spec and compare. Do not copy a random clone onto a client USB and call it the audit kit.

What it finds

  • Browser history — visits to 80+ known AI services (ChatGPT, Claude, Gemini, Copilot, Grok, DeepSeek, Perplexity, and more), with visit counts and timestamps. Chrome, Edge, Firefox, Brave, Arc, Safari, Chromium — including Linux snap/flatpak profiles.
  • Installed software — AI desktop apps and AI-enabled tools (dpkg/rpm on Linux, registry inventory via PowerShell on Windows, /Applications on macOS).
  • DNS logs — Pi-hole, OPNsense, dnsmasq, or syslog query logs matched against the AI domain watchlist.
  • Risk ranking — every finding scored CRITICAL/HIGH/MEDIUM/LOW, with a PIPEDA/PHIPA exposure flag for Canadian privacy compliance.

Output

Each scan produces an audit-evidence package:

File Purpose
report.html Branded, print-ready findings report
report.pdf PDF version (when WeasyPrint is available)
report.json Structured findings for tooling
inventory.csv Flat spreadsheet of findings
findings.db SQLite database of findings
machine_info.json Host metadata (hostname, user, OS)
scanner.log Timestamped activity log — the audit trail

Quick start

Windows: double-click run.bat (or run.bat --express --client "Acme Corp"). No Python installation required — the kit can run from a bundled portable Python on a USB stick.

macOS: double-click run.command, or bash run.sh in Terminal. Note: Safari history requires Full Disk Access for your terminal — the scanner detects this and tells you exactly what to grant instead of crashing.

Linux: bash run.sh or python3 audit-kit.py.

Two modes:

  • Wizard (default) — interactive 5-step flow: client info, scan mode, output location, confirm, run. This is the field-audit experience.
  • Express./audit-kit.py --express --client "Acme" --auditor "Jane". No prompts; auto-detects everything.
  • Dry run./audit-kit.py --dry-run. Lists every file and directory the scanner would read on this machine, then exits without reading anything. Use it to confirm scope with the client before touching data on privacy-sensitive engagements (PIPEDA/PHIPA).

Targeted scans via environment variables: AUDITKIT_SKIP_BROWSER=1 or AUDITKIT_SKIP_SOFTWARE=1.

The watchlist

ai_domains.json is the curated AI-service watchlist — the single source of truth for what counts as a finding. Entries carry a default risk level, category, and recommended action. PRs welcome: new AI services launch weekly, and keeping this list current is a community job. Entries with scan: false are infrastructure/cloud domains tracked for reference but not reported as Shadow AI findings.

Tests

python3 -m pytest tests/ -v

27 tests covering the domain matcher, software inventory parsers, browser path detection (including snap and Windows Store-stub edge cases), and regressions from real field runs. tests/fixtures-browser-matrix/ carries synthetic Chrome/Edge/Brave/Firefox history databases (plus a Safari fixture) used for cross-browser regression runs on all three platforms.

Building standalone binaries

The kit can ship as a single-file executable (no Python needed on the target):

pip install pyinstaller
pyinstaller scanner.spec

Cross-platform binaries built and verified live from this code: Linux x64, Windows 11 x64 (QEMU/KVM), and macOS arm64 (26.3) — see scanner.spec for the build recipe.

Run it yourself — or hire us

This kit is the same tool elect-rix uses on paid Shadow AI Discovery audits. Running it yourself gives you the data. A professional audit gives you the interpretation, the remediation roadmap, the staff policy templates, and a report your insurer and counsel will accept.

elect-rix Technology Solutions — sovereign AI infrastructure for Canadian business. Your data stays in your building.

License

MIT (see LICENSE). Code and watchlist are free to use, fork, and extend. The elect-rix name and branding are trademarks of RixBot Technologies Inc. and not licensed for commercial reuse — rebrand the report template if you redistribute.

About

Find which AI tools are actually in use on a machine — from a USB stick. Cross-platform Shadow AI Discovery scanner producing risk-ranked audit reports. No cloud, no telemetry. MIT licensed.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages