Safeguard
Vulnerability Analysis•

CVE-2026-32052: OpenClaw system.run Hidden Command Execution

CVE-2026-32052 lets hidden positional arguments run under misleading approval text in OpenClaw system.run. Upgrade to OpenClaw 2026.2.24 or later.

Safeguard Security Team
2 min read

CVE-2026-32052 is a command injection in the OpenClaw system.run shell wrapper, scored 6.4 by NVD. Extra positional arguments after an inline shell payload could run while the approval text showed something else. Upgrade OpenClaw to 2026.2.24 or later. Checks and steps follow below.

What's affected

OpenClaw 2026.2.23 and earlier. The patched version is 2026.2.24.

It concerns approval of commands in system.run, where a person or policy approves what the screen shows before the command runs. The user sees one command and the system runs another.

How to check if you're exposed

OpenClaw is distributed on npm, so read the installed version.

npm ls openclaw
npm ls -g openclaw

OpenClaw release numbers are date based, so compare the version against the fixed release named above. If you run it in a container, check the image you deploy and rebuild it after updating.

What to do now

  1. Upgrade OpenClaw to 2026.2.24 or later.
  2. Until you do, treat approvals of shell wrapper commands as unreliable, and avoid approving commands you cannot read in full.
  3. Review your allowlist and approval policy for shell wrappers.
  4. Check history for system.run calls that you do not recognize.

Why it matters

The record lists CWE-436, interpretation conflict, and CWE-77, command injection. The display text and the executed argument list were built from different views of the same command. When they disagree, an approval is only as good as the text it was shown.

The fix binds the approval text to the full formatted argument list and rejects mismatches. It also validates consistency in the macOS companion app.

The record does not flag this CVE as listed in CISA's Known Exploited Vulnerabilities catalog, and no advisory I read says it is exploited in the wild. Treat it as a patch to schedule promptly, not a fire drill, unless your own logs say otherwise.

Quick check

What is the core lesson?

An approval step must show exactly what will run. If the display and execution differ, the check is worthless.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.