OpenClaw Security: What You Must Configure Before You Trust It
The checklist to work through before connecting real accounts
By Tal SlavinUpdated
Is OpenClaw safe to use?
OpenClaw runs with broad system access and sandboxing turned off by default, so treat the Gateway as a trusted process that can execute untrusted tool calls. Before trusting it, review tool permissions, enable sandboxing, protect API keys, limit which channels can trigger risky actions and keep it updated.
☰On this page (2 sections)
OpenClaw runs with full system access by default and sandboxing turned off. Treat the Gateway as a trusted process that can execute untrusted tool calls. If you have not installed it yet, see how to install OpenClaw.
Immediate actions#
- Review tool permissions and disable anything unnecessary.
- Enable sandboxing for non-main sessions where possible.
- Store API keys carefully (avoid plain-text config when feasible).
- Limit which channels can trigger high-privilege actions.
- Keep the Gateway updated.
- Use a dedicated machine or VM for production.
- Never expose the Control UI to the public internet without authentication.
Third-party skills deserve the same scrutiny: review a skill's permissions before enabling it (OpenClaw skills and ClawHub). On a shared team Gateway, restrict who can trigger high-privilege tools (OpenClaw for teams).
The OpenClaw Foundation and community publish security guidance. Follow it before giving the agent access to email, payments or production systems. The general risks, prompt injection above all, are covered in the best super intelligence for autonomous cybersecurity.
Sources & Further Reading#
| Claim / Topic | Source | Date |
|---|---|---|
| Sandboxing off by default; trusted Gateway, untrusted execution | OpenClaw docs | Read 2026-11-03 |
Frequently asked questions
Is OpenClaw safe to use?
OpenClaw runs with broad system access and sandboxing turned off by default, so treat the Gateway as a trusted process that can execute untrusted tool calls. Before trusting it, review tool permissions, enable sandboxing, protect API keys, limit which channels can trigger risky actions and keep it updated.
Is OpenClaw sandboxed by default?
No. Sandboxing is available but turned off by default. Enable it for non-main sessions where possible, and use a dedicated machine or VM for production.
Should I expose the OpenClaw Control UI online?
Never expose the Control UI to the public internet without authentication. Anyone who reaches it could direct an agent that can run commands on your machine.
How we sourced this. OpenClaw docs read on 2026-11-03. Methodology