OpenClaw Security: What You Need to Know

The Double-Edged Sword of AI Agents

OpenClaw represents a new paradigm in personal computing — an AI agent with deep system access that can execute tasks autonomously. This power comes with inherent security risks that every user must understand.

Unlike traditional chatbots that only process text, OpenClaw can:

• Execute commands on your computer
• Access your files and data
• Control web browsers
• Send messages on your behalf
• Install and run arbitrary code (skills)

Known Vulnerabilities

WebSocket Hijacking (CVE Pending) — Patched in v2026.1.29

In February 2026, security researchers discovered a critical vulnerability in OpenClaw's WebSocket implementation[1]. An attacker could:

• Hijack OpenClaw sessions remotely
• Execute arbitrary commands on affected systems
• Access sensitive data processed by OpenClaw

Severity: HIGH — This vulnerability affected over 40,000 exposed instances before patching[2].

Token Theft via Malicious Skills

Research found approximately 17% of OpenClaw "Skills" contained malicious code[3]. Common malware types included:

• AMOS Stealer — credential theft
• Token harvesters — API key extraction
• Backdoors — persistent remote access

Government Warnings

Security concerns have reached government level:

• China (February 5, 2026): Issued official security warnings about OpenClaw, citing risks of data theft and system control[4]
• Kaspersky (February 2026): Declared OpenClaw "not safe for enterprise use" without proper security controls[2]

Why OpenClaw Has These Risks

The security challenges stem from OpenClaw's core design:

Required Permissions

OpenClaw needs extensive system access to function:

• File system read/write access
• Ability to spawn processes
• Network access for API calls
• Browser automation via CDP

Unverified Skills Ecosystem

Anyone can publish OpenClaw skills. While the community tries to vet submissions, malicious actors can:

• Obfuscate malicious code
• Use typosquatting on popular skill names
• Release clean versions then add malware later

How to Run OpenClaw Safely

1. Always Update to Latest Version

Check your version regularly:

Update immediately when new releases are available:

2. Only Install Verified Skills

• Review skill code before installation
• Stick to official or well-known community skills
• Check skill repositories for activity and reviews
• Be suspicious of skills requesting unnecessary permissions

3. Run in a Sandbox/VM (Advanced Users)

For maximum security, run OpenClaw in:

• A dedicated virtual machine
• A container with restricted network access
• A separate user account with limited permissions

4. Use API Keys Wisely

• Create separate API keys for OpenClaw
• Set spending limits on your AI provider account
• Rotate keys periodically
• Never commit API keys to repositories

5. Monitor Activity

OpenClaw provides session monitoring:

Review logs regularly for suspicious activity.

OpenClaw's Security Improvements

Following the February 2026 revelations, the OpenClaw team has implemented:

• Enhanced skill scanning: Automated analysis of published skills for malware patterns[5]
• Signed skills: Cryptographic verification of skill authors
• Security advisories: Regular bulletins for newly discovered issues

Should You Use OpenClaw?

The answer depends on your use case and risk tolerance:

Conclusion

OpenClaw offers powerful capabilities but requires responsible security practices. By keeping software updated, vetting skills carefully, and understanding the risks, users can safely explore the future of personal AI agents.

The security landscape is evolving rapidly. Stay informed by following @openclaw on X and monitoring official security channels.

References

1. Security Research Report - WebSocket Vulnerability Analysis, February 2026
2. Kaspersky Official Blog - "OpenClaw: 40,000+ Exposed Instances" - February 2026
3. Malware Analysis - "AMOS Stealer Distribution via OpenClaw Skills" - February 2026
4. Reuters - "China warns of security risks linked to OpenClaw" - February 5, 2026
5. OpenClaw v2026.2.6 Release Notes - Enhanced security scanner for malicious skills