GHSA-8m9v-xpgf-g99m: OpenClaw has an unauthorized sender bypass in its stop triggers and /models command authorization
OpenClaw up to 2026.2.26 skipped sender authorization on two command paths: stop-like natural-language abort triggers and the `/models` command. An unauthorized sender could interrupt active sessions and read model and auth metadata. Upgrade the npm package `openclaw` to 2026.3.1.
Published Mar 2, 2026 · Analysis Oct 31, 2026
What an attacker can do
OpenClaw takes messages from senders and treats some as commands. Most commands check that the sender is authorized. Two did not. Phrases that read as a request to stop, which the agent treats as an abort trigger, worked for anyone who could message the agent. The /models command also answered any sender.
The first path gives an unauthorized sender integrity impact on availability of work: they can end a running agent session mid-task. The second leaks metadata about the configured models and their authentication that the operator meant to gate behind authorization. The advisory does not list which fields /models returns, so treat provider names and auth state as exposed.
The attacker needs no account or privileges, only the ability to send a message through a channel the agent listens on. The advisory does not describe code execution or access to files. Its CVSS 4.0 vector scores integrity impact as low and the rest as none.
Who is affected
The npm package openclaw at version 2026.2.26 and earlier is affected; the advisory data lists everything below 2026.3.1. Any deployment where people other than authorized users can message the agent is exposed, such as shared chat channels or group chats.
Check the installed version:
npm ls openclawor run openclaw --version.
openclaw
- Ecosystem
- npm
- Affected
- < 2026.3.1
- Fixed in
- 2026.3.1
- Check
npm ls openclaw
How to fix it
Upgrade to openclaw 2026.3.1. From that release, sender authorization is enforced for stop-like abort triggers and for /models listings.
npm install -g [email protected]The advisory gives no workaround. Until you upgrade, restrict who can message the agent to trusted senders.
How to tell if you were hit
Review message and session logs for stop-like messages or /models requests from senders who are not on your authorized list. Look for active sessions that ended early right after a message from an unknown sender. The advisory names no specific log line. With Beam, recorded agent activity can show which sessions stopped unexpectedly, though not who sent the abort message.
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
Summary
Unauthorized senders could trigger two command paths without sender authorization checks:
- stop-like natural-language abort triggers
/modelscommand output
Impact
An unauthorized sender could disrupt active sessions and view model/auth metadata that should be authorization-gated.
Fix
Sender authorization is now enforced for stop-like abort triggers and /models listings.
Affected and Patched Versions
- Affected:
<= 2026.2.26 - Patched:
2026.3.1
Credited to tdjackey.
Questions
- Is this the same as a remote code execution bug?
No. The advisory describes two missing authorization checks. The impact is disrupted sessions and exposed model and auth metadata, not code execution.
- Does this need authentication?
No. The CVSS 4.0 vector has no privileges required and no user interaction. Any sender who can reach the agent's message channel can trigger it.
- Does this advisory have a CVE?
No. It is tracked only as GHSA-8m9v-xpgf-g99m, published on 2026-03-02 and rated medium with CWE-863. It is not in CISA's KEV catalog.
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 entriesCVE-2026-62197
OpenClaw: Browser CDP discovery could accept blocked WebSocket URLs
medium8.5Oct 21, 2026CVE-2026-35630
OpenClaw: QQBot native approval buttons did not enforce configured approver identity
high8.0Jul 2, 2026CVE-2026-53806
OpenClaw: Combined POSIX shell options could confuse exec revalidation
high8.8Jul 2, 2026CVE-2026-53809
OpenClaw: Embedded runner policy could be confused by provider aliases
medium3.8Jul 2, 2026CVE-2026-53810
OpenClaw's marketplace runtime extension metadata could point at unscanned payloads
high8.8Jul 2, 2026CVE-2026-53811
OpenClaw: Matrix allowFrom could bind to mutable display names
high8.8Jul 2, 2026
