🛡️ Security Checklist
Essential hardening for your OpenClaw installation
OpenClaw instances detected on the public internet — many with security issues
🔔 Latest: Update to 2026.2.1
Released Feb 2, 2026 — includes multiple security fixes:
- Path traversal fixes — WhatsApp accountId, MEDIA path extraction, message-tool filePath
- LFI prevention — Restricted local path extraction in media parser
- Exec hardening — Blocked LD*/DYLD* env overrides for host exec
- Browser relay secured — Chrome extension CDP sessions now protected
- TLS 1.3 minimum — Now required for TLS listeners
npm update -g openclaw && ocl gateway restart
📰 Coverage: The Hacker News · The Register
Gateway Binding CRITICAL
By default, OpenClaw binds to all interfaces (0.0.0.0), making it accessible from the internet if your firewall allows it.
Bind to loopback only
Unless you need remote access, bind to localhost so only local connections work.
# In your config.yaml or via CLI: gateway: bind: "0.0.0.0" # ❌ BAD - exposed to network bind: "loopback" # ✅ GOOD - localhost only bind: "127.0.0.1" # ✅ GOOD - same as loopback
Check current binding
Run this to see what your gateway is bound to:
openclaw config get gateway.bind
# Or check what's listening:
lsof -i :3000 | grep LISTEN
netstat -an | grep 3000
Authentication CRITICAL
Control who can interact with your gateway and through which channels.
Set DM policy to "pairing"
Require manual approval before new devices/users can interact.
# In config.yaml: gateway: dmPolicy: "pairing" # ✅ Requires approval dmPolicy: "open" # ❌ Anyone can message
Whitelist allowed user IDs
Only allow specific user IDs to interact with your bot.
# In your channel config: telegram: allowedUsers: - "123456789" # Your Telegram user ID - "987654321" # Family member, etc.
Set a gateway token
If exposing API endpoints, require authentication.
gateway:
token: "your-secret-token-here"
Tool & Shell Access HIGH
Control what the AI can do on your system.
Enable sandbox mode for group chats
Limit tool access when in shared/group contexts.
# In AGENTS.md, add guidelines:
## Group Chat Rules
- Never run shell commands in group chats
- No file access outside designated folders
- Ask before any external actions
Review tool allowlist
Check which tools are enabled and disable any you don't need.
openclaw config get tools # Disable dangerous tools if not needed: tools: exec: false # No shell commands browser: false # No browser control
Use confirmation prompts for destructive actions
Configure OpenClaw to ask before deleting files or running certain commands.
Network Security HIGH
Secure your network access properly.
Use Tailscale or Cloudflare Tunnel for remote access
Never expose your gateway port directly to the internet. Use a secure tunnel instead.
# ❌ DON'T: Open port 3000 on your router # ❌ DON'T: Use ngrok without auth # ✅ DO: Use Tailscale (encrypted mesh VPN) tailscale up # Access via: http://100.x.x.x:3000 (Tailscale IP) # ✅ DO: Use Cloudflare Tunnel cloudflared tunnel --url http://localhost:3000
Check firewall rules
Ensure your firewall isn't allowing external access to the gateway port.
# macOS: sudo pfctl -sr | grep 3000 # Linux (ufw): sudo ufw status # Linux (iptables): sudo iptables -L -n | grep 3000
API Keys & Credentials HIGH
Protect your API keys and credentials.
Store API keys securely
Use environment variables or the system keychain, not plaintext files.
# ❌ DON'T: Put keys in config.yaml anthropic: apiKey: "sk-ant-..." # Bad! # ✅ DO: Use environment variables export ANTHROPIC_API_KEY="sk-ant-..." # ✅ DO: Use macOS Keychain security add-generic-password -a $USER -s anthropic-api-key -w "sk-ant-..."
Use scoped API keys where possible
Create dedicated API keys with limited permissions for OpenClaw.
Rotate keys if compromised
If you suspect your OpenClaw was accessed, rotate all API keys immediately.
Skill Security MEDIUM
Be careful when installing third-party skills.
Audit skills before installing
Read ALL files in a skill, not just SKILL.md. Look for hidden scripts, curl commands, or external URLs.
# After installing, check the skill folder: ls -la ~/.openclaw/skills/[skill-name]/ cat ~/.openclaw/skills/[skill-name]/* # Look for red flags: grep -r "curl\|wget\|bash -c" ~/.openclaw/skills/[skill-name]/
Check skill source/author
Only install skills from trusted sources. Don't trust download counts — they can be faked.
🔍 Quick Security Audit
Run this command to check your current security status:
openclaw doctor --security
# Or manually check key settings:
echo "=== Gateway Binding ==="
openclaw config get gateway.bind
echo "=== DM Policy ==="
openclaw config get gateway.dmPolicy
echo "=== Listening Ports ==="
lsof -i -P | grep openclaw
Related Guides
- Tailscale Setup — Secure remote access
- Cloudflare Tunnel — Alternative to Tailscale
- Official Security Docs