← back to Claw Guide
Claw Guide · research

OpenClaw security checklist: browsers, credentials, skills, and sandboxing

OpenClaw’s defaults are conservative for a single operator on a laptop. The trouble starts the moment you point it at a server, a shared machine, or a customer’s data — because two of the protections you would assume are on, are not. Sandboxing and exec approvals ship off by default, and that is the project saying so, not a critic.

The short answer. Run the built-in audit first, then work the four things that actually move the needle: one trust boundary per gateway, the gateway kept off the public internet, the sandbox turned on, and secrets moved out of plain text. Everything else is follow-up.

Why the defaults are not enough

The evidence is public and specific. Cisco’s AI security team tested a third-party OpenClaw skill and found it performing data exfiltration and prompt injection without the user’s awareness, noting the skill repository lacked the vetting to stop malicious submissions. Reviewing OpenClaw 2.0, The Register reported that the Secret Store is not encrypted at rest and the sandbox is not enabled by default, concluding the release “is not bringing security by default.” And OpenClaw’s own docs are blunt about the scope: default OpenClaw “is a trusted single-operator assistant,” and one gateway is one trust domain — not a boundary between people who do not trust each other.

The checklist

  1. Run the audit before you change anything

    OpenClaw ships its own audit. Start there — it reports what you have drifted from and, for some checks, offers an auto-fix.

    openclaw security audit
    openclaw security audit --deep   # schedule this, and alarm on the check IDs
  2. Keep one trust boundary per gateway

    Do not run mutually distrusting people on one gateway. For mixed-trust use, split the boundary: separate gateway, separate credentials, ideally separate OS users or hosts. This is a design constraint, not a config flag.

  3. Keep the Gateway off the public internet

    On a normal host install the Gateway binds to loopback. Container images are the exception — they default to an exposed bind, so pair that with auth. For remote access, front it with Tailscale or an identity-aware proxy rather than opening a port.

  4. Leave DM policy on pairing

    By default an unknown sender gets a pairing code instead of a processed message, and groups are allowlisted behind a mention gate. Keep that. “Open” DMs on a gateway that touches email and files is how strangers end up driving your agent.

  5. Turn the sandbox on

    This is the big one, because it is off by default and it is the layer that contains a compromised run:

    agents.defaults.sandbox.mode: "all"
    agents.defaults.sandbox.workspaceAccess: "ro"   # unless the agent owns the workspace

    Run it on the openshell or docker backend. Sandboxed execution defaults to no network (network: "none") or a default-deny policy.

  6. Bound what tools and exec can do

    Pick a permission mode per session — guarded or workspace. full requires operator.admin; do not hand it out casually. Give each agent the narrowest profile that still does its job.

  7. Move secrets into SecretRefs

    Plain-text keys in a config or a workspace .env are exposed the moment the host is. Reference secrets instead, and check them in CI:

    openclaw secrets audit --check
  8. Treat every third-party skill and plugin as untrusted code

    The Cisco finding is the reason. ClawHub shows scan status, but a pending or stale scan can still let an install through — and native plugins run in-process, not in the sandbox. Pin versions, lock dependencies, allowlist what you install, and verify before adding:

    openclaw skills verify
  9. Watch the browser

    A real browser is a real attack surface. OpenClaw’s browser SSRF policy is strict by default — keep it, and keep the bypass flags off in production.

  10. Monitor and patch like a server

    Turn on message auditing, export OpenTelemetry to your SIEM with operator-owned retention, schedule the deep audit, and treat security advisories as a standing feed. The project ships CVEs; you are the one who applies them.

  11. Have a one-page incident plan

    The project publishes the order: contain (isolate the instance), rotate (assume any reachable secret is compromised), audit (read the logs and transcripts), then collect evidence before you wipe anything.

The boundary. This is hardening guidance for a deployment you run, assembled from OpenClaw’s public documentation and independent reporting — not a claim that any setup is unbreakable. It cannot make an agent safe against a determined adversary, and it does not cover hosted providers, whose internal security you cannot inspect from the outside. The honest position, in the project’s own words: sandboxing and exec approvals are off by default, so safety is something you configure, not something you install.
Get the weekly OpenClaw Ops Brief.
Deployment notes, workflow tests, and agent-stack updates — one email a week. Subscribe on the homepage.
by Jonas Reyes — the builder’s desk, Side Quest Studios
AI-assisted, curated for Side Quest Studios.

References

  1. OpenClaw documentation — Security (trust model, defaults, audit): docs.openclaw.ai/gateway/security
  2. — Hardened baselines: docs.openclaw.ai/gateway/security/hardened-baseline
  3. — Security audit checks: docs.openclaw.ai/gateway/security/audit-checks
  4. — Secrets and storage: docs.openclaw.ai/gateway/secrets
  5. — “Why OpenClaw” (trust boundary, what the project does not claim): docs.openclaw.ai/start/why-openclaw
  6. — Published security advisories: github.com/openclaw/openclaw/security/advisories
  7. Cisco AI security — “Personal AI Agents like OpenClaw Are a Security Nightmare”: blogs.cisco.com
  8. The Register — “OpenClaw 2.0 pours glitter on slow-burning security dumpster fire”: theregister.com

Claw Guide is published by the team behind Claw Way, a managed OpenClaw host. We disclose that on every page. This checklist is written for self-hosters and applies to any deployment; it deliberately does not describe how our own platform is built. · contact