OpenClaw Security Best Practices: Keeping Your AI Assistant Safe

2026-02-02 · Nia

Running an AI assistant that can access your files, execute commands, and connect to your accounts is powerful—but with great power comes great responsibility. Here's how to keep your OpenClaw installation secure.

Why Security Matters for AI Assistants

Unlike traditional chatbots, OpenClaw has real capabilities:

File system access — Can read and write files
Command execution — Can run shell commands
API integrations — Connects to your services
Message access — Reads your communications

This makes security not optional, but essential.

1. API Key Management

Never Hardcode API Keys

❌ Wrong:

anthropic:

apiKey: "sk-ant-api03-xxxxx"

✅ Right:

anthropic:
apiKey: "${ANTHROPIC_API_KEY}"

Store keys in environment variables or secure files:

# Create secure config directory mkdir -p ~/.config/openclaw chmod 700 ~/.config/openclaw Store API key echo "your-api-key" > ~/.config/openclaw/anthropic_key

chmod 600 ~/.config/openclaw/anthropic_key

Rotate Keys Regularly

Set calendar reminders to rotate API keys quarterly. If you suspect compromise, rotate immediately.

Use Separate Keys for Development/Production

Don't use your production API keys for testing. Create separate keys with appropriate rate limits.

2. Access Control

Restrict Owner Numbers

Only allow trusted phone numbers to control OpenClaw:

ownerNumbers:

- "1234567890" # Your primary number only

Never add numbers you don't control. Each owner has full access.

Channel-Specific Permissions

Configure different permission levels per channel:

telegram: # Private chat - full access privateChat: allowCommands: true allowFileAccess: true # Group chats - limited access groupChats: allowCommands: false

allowFileAccess: false

Implement Command Allowlists

Restrict which shell commands OpenClaw can execute:

exec: security: allowlist allowlist: - "git" - "npm" - "python3"

# Don't add rm, sudo, etc.

3. File System Security

Use Workspace Isolation

Keep OpenClaw confined to its workspace:

workspace: /home/openclaw/.openclaw/workspace

Don't give it access to system directories or sensitive files.

Implement Read-Only Paths

Mark sensitive directories as read-only:

filesystem:
readOnly:
- ~/.ssh
- ~/.gnupg
- /etc

Never Store Secrets in Workspace

Your workspace files may be referenced in conversations. Never store:

Passwords
Private keys
API tokens
Personal documents

Use separate, permission-protected directories for sensitive data.

4. Network Security

Run Behind a Firewall

If self-hosting, restrict network access:

# Allow only necessary outbound connections ufw default deny outgoing ufw allow out 443/tcp # HTTPS

ufw allow out 53/udp # DNS

Use HTTPS for Webhooks

Always use HTTPS for webhook endpoints:

telegram:

webhookUrl: "https://your-domain.com/webhook" # Not http://

Implement Rate Limiting

Protect against abuse:

rateLimit: messagesPerMinute: 20

commandsPerMinute: 10

5. Monitoring and Logging

Enable Audit Logging

Log all commands and file operations:

logging: level: info auditCommands: true auditFileAccess: true

logFile: /var/log/openclaw/audit.log

Set Up Alerts

Get notified of suspicious activity:

alerts: onFailedAuth: true onUnknownNumber: true

onHighUsage: true

Regular Log Review

Schedule weekly log reviews to catch anomalies:

# Check for failed authentication attempts grep "auth failed" /var/log/openclaw/audit.log Check for unusual commands

grep "exec:" /var/log/openclaw/audit.log | sort | uniq -c | sort -rn

6. Data Privacy

Configure Memory Carefully

Be thoughtful about what OpenClaw remembers:

memory: enabled: true excludePatterns: - "password" - "secret"

- "api_key"

Implement Data Retention

Don't keep data forever:

retention: conversationHistory: 30 # days

fileCache: 7 # days

Separate Personal and Shared Contexts

As noted in customizing OpenClaw, keep personal context separate from shared spaces:

Load MEMORY.md only in private sessions
Don't leak personal info in group chats

7. Update and Patch Management

Keep OpenClaw Updated

Regular updates include security patches:

# Check for updates openclaw update check Apply updates

openclaw update run

Subscribe to Security Announcements

Follow the OpenClaw GitHub for security advisories.

Test Updates Before Production

Always test updates in a staging environment first.

8. Backup and Recovery

Regular Backups

Back up your configuration and workspace:

# Daily backup script
tar -czf openclaw-backup-$(date +%Y%m%d).tar.gz \
~/.openclaw/workspace \
~/.config/openclaw

Secure Backup Storage

Encrypt backups before storing:

gpg --encrypt --recipient your@email.com openclaw-backup.tar.gz

Test Recovery

Periodically test that you can restore from backups.

Security Checklist

Use this checklist for your OpenClaw installation:

[ ] API keys stored securely (not in code)
[ ] Owner numbers restricted to trusted contacts
[ ] Workspace isolated from system files
[ ] Sensitive directories marked read-only
[ ] HTTPS enabled for all webhooks
[ ] Rate limiting configured
[ ] Audit logging enabled
[ ] Regular updates scheduled
[ ] Backups automated and tested

Common Security Mistakes

Mistake 1: Public Workspace

Never make your workspace publicly accessible. It contains conversation history and potentially sensitive operations.

Mistake 2: Overly Permissive Commands

Don't allow dangerous commands:

# DANGEROUS - don't do this exec:

security: full # Allows any command including rm -rf /

Mistake 3: Sharing Bot Tokens

Never share your Telegram bot token or other channel credentials. Anyone with the token can impersonate your bot.

Mistake 4: Ignoring Updates

Security vulnerabilities are discovered regularly. Running outdated software is one of the top security risks.

Conclusion

Security isn't a one-time setup—it's an ongoing practice. By following these best practices, you can enjoy OpenClaw's powerful capabilities while keeping your data and systems safe.

For more on setting up OpenClaw, see our getting started guide. To learn about automation features, check out cron jobs and scheduled tasks.

Stay secure, stay productive. More OpenClaw guides at blog.youmake.dev.