mac-security-audit

Security audit and hardening script for macOS machines running AI agents — OpenClaw, Ollama, Open WebUI, LM Studio, and more.

Built for Mac Minis and other Macs used as always-on AI workstations. One command to find what's exposed, what's misconfigured, and what needs locking down.

Why this exists

Mac Minis are becoming the go-to for local AI — Ollama, OpenClaw, Open WebUI, LM Studio. But most setups skip security basics:

• Ollama bound to 0.0.0.0 (your models are accessible to the whole network)
• OpenClaw running with weak auth tokens and KeepAlive daemons
• Docker containers exposing ports on all interfaces
• Tailscale Funnel accidentally publishing local services to the internet
• Orphaned agent configs with API keys and secrets sitting on disk
• Firewall off, FileVault off, sleep disabled with no auto-restart

This script finds all of that in seconds.

Quick start

Audit only (safe, read-only):

curl -fsSL https://raw.githubusercontent.com/fullstacksushil/mac-security-audit/main/mac-security-audit.sh | bash

Full audit (includes firewall and sharing checks):

curl -fsSL https://raw.githubusercontent.com/fullstacksushil/mac-security-audit/main/mac-security-audit.sh | sudo bash

Audit + auto-fix (fixes what it can):

curl -fsSL https://raw.githubusercontent.com/fullstacksushil/mac-security-audit/main/mac-security-audit.sh | sudo bash -s -- --fix

Or clone and run:

git clone https://github.com/fullstacksushil/mac-security-audit.git
cd mac-security-audit
chmod +x mac-security-audit.sh
sudo ./mac-security-audit.sh --fix

What it checks

macOS Security

Check	Details	Auto-fix
Firewall	Global state + stealth mode	Yes
FileVault	Disk encryption	Manual
SIP	System Integrity Protection	Manual
Gatekeeper	App verification	Yes
Remote access	SSH, Screen Sharing, ARD	Yes (SSH)
Home directory	Should be chmod 700	Yes
SSH keys	Permissions on private keys	Yes
Software updates	Pending macOS updates	Manual

AI Agent Security

Check	Details	Auto-fix
OpenClaw gateway	Running status, PID, memory	-
OpenClaw auth	Token strength (flags weak tokens)	Manual
OpenClaw binding	Loopback vs all-interfaces	Manual
OpenClaw plugins	Unvetted plugin detection	Manual
OpenClaw LaunchAgent	KeepAlive, auto-restart config	Manual
Orphaned agents	Detects leftover secrets from unused agents	Manual
Ollama binding	OLLAMA_HOST in shell config + runtime	Yes
Docker containers	Ports exposed on 0.0.0.0	Manual
Docker daemons	Orphaned LaunchDaemons	Yes
All AI processes	Ollama, LLaMA, LM Studio, ComfyUI, etc.	-

Network & Services

Check	Details	Auto-fix
Listening ports	All TCP listeners with wildcard warnings	-
Tailscale	Serve/Funnel exposure detection	Manual
LaunchAgents	Custom user agents (filters Apple/vendors)	-
LaunchDaemons	Third-party system daemons	-
Cron jobs	User crontab entries	-
Login items	Apps that start on login	-
Sensitive files	.env, credentials.json, .netrc, etc.	-

Server Configuration

Check	Details	Auto-fix
Sleep settings	Reports if sleep is off (expected for servers)	-
Auto-restart	Restart on power failure	Yes
Power management	Display sleep, hibernate settings	-

Sample output

  ┌─────────────────────────────────────────┐
  │       mac-security-audit v1.1.0          │
  │   Security audit for macOS AI stations   │
  └─────────────────────────────────────────┘

  Mode: AUDIT + FIX

═══ Firewall ═══
[ OK ] Firewall is enabled
[ OK ] Stealth mode is enabled

═══ AI Agent Services ═══
[INFO] OpenClaw gateway is RUNNING (PID 1401, 1.0% RAM)
[WARN] OpenClaw LaunchAgent installed
[ALERT] OpenClaw gateway token is weak (6 chars) — use a strong secret

═══ Ollama ═══
[ OK ] Ollama is running
[ALERT] Ollama bound to 0.0.0.0 in /Users/you/.zshrc — exposes API to network
[FIXED] Changed OLLAMA_HOST to 127.0.0.1 in /Users/you/.zshrc

═══ Audit Complete ═══
  Found 2 critical issue(s) that should be addressed.
  Auto-fixed 1 issue(s).

Output legend

Tag	Meaning
[ OK ]	Check passed
[INFO]	Informational — review manually
[WARN]	Non-critical but worth reviewing
[ALERT]	Critical issue — should be fixed
[FIXED]	Issue auto-fixed (--fix mode)

Requirements

• macOS 12+ (Monterey or later)
• Bash 3.2+ (ships with macOS)
• No dependencies — uses only built-in macOS tools

Contributing

Found something else that should be checked? Open an issue or PR. Especially interested in:

• Additional AI agent frameworks to detect
• macOS hardening checks specific to headless/server use
• Homebrew formula for easier installation

License

MIT