NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · OpenClaw

GHSA-6x2m-hqfw-hvpj: OpenClaw: Node exec approvals could be replayed across nodes

mediumCVSS 4.0 6.9

In OpenClaw 2026.2.22-2 and earlier, `exec.approval` requests for `host=node` were not bound to the target `nodeId`, so an operator's approval of a `system.run` request for one node could be replayed on a different node in the same gateway fleet. Fixed in 2026.2.23.

Published Mar 2, 2026 · Analysis Oct 31, 2026

What an attacker can do

When an agent wants to run a command on a remote node, the operator approves a system.run request. The approval record did not carry the node identity. Execution checks therefore had nothing to compare against the node that was actually invoked. An approval given for the command on node A would also satisfy the same request on node B.

The damage is limited to the operator-controlled fleet behind one gateway. Someone who can submit requests, such as an agent redirected by a prompt injection, can run an already approved command on a node where the operator never agreed to it. If the nodes differ in what they hold, for example a build host and a production host, the consequences differ with them. The registry rates the integrity impact low. The advisory reports no confidentiality or availability effect and no way around the approval for a command that was never approved.

Who is affected

The openclaw npm package at 2026.2.22-2 and earlier (below 2026.2.23), in fleets with more than one paired node where system.run approvals are used.

npm ls -g openclaw
openclaw --version

openclaw

Ecosystem
npm
Affected
< 2026.2.23
Fixed in
2026.2.23
Check
npm ls openclaw

How to fix it

Upgrade to openclaw 2026.2.23.

npm install -g [email protected]

The fix commit 4a3f8438e527ac371a67fe7ac68a287f0dbe6063 requires and stores nodeId for host=node approval requests, and rejects execution when the approving node binding does not match the invoking node. The advisory names no workaround.

How to tell if you were hit

Compare approval records with execution logs: the same system.run command executed on several nodes after a single approval is the signature. Look for commands that ran on a node other than the one named in the approval prompt. Beam records agent shell commands per host, which makes it easier to see where an approved command also ran.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

Summary

exec.approval requests for host=node were not explicitly bound to the target nodeId, so an approval intended for one node could be replayed for a different node under the same operator-controlled gateway fleet.

Impact

An operator approval for a system.run request could be reused across nodes if the request payload did not carry node identity through approval and execution checks.

Affected Packages / Versions

  • Package: openclaw (npm)
  • Affected: <= 2026.2.22-2
  • Fixed: 2026.2.23 (released)

Mitigation

Upgrade to 2026.2.23 or later once published.

Fix Details

The fix requires and persists nodeId for host=node approval requests and rejects execution when the approving node binding does not match the invoking node.

Fix Commit(s)

  • 4a3f8438e527ac371a67fe7ac68a287f0dbe6063

Release Process Note

patched_versions is pre-set to the released version (2026.2.23). This advisory now reflects released fix version 2026.2.23.

OpenClaw thanks @tdjackey for reporting.

Credited to tdjackey.

Questions

What does replay across nodes mean here?

An approval given for a system.run request on one node was accepted for the same request on another node, because the approval carried no node identity.

Does this bypass approval entirely?

No. The command still needed an approval. The flaw is that the approval was not tied to the node it was given for.

Which version fixes GHSA-6x2m-hqfw-hvpj?

OpenClaw 2026.2.23, which requires nodeId on approvals and rejects a mismatch.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More OpenClaw vulnerabilities

All OpenClaw entries