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

GHSA-7jm2-g593-4qrc: OpenClaw: Agent gateway config mutations could change protected operator settings

mediumCVSS 4.0 6.0

In OpenClaw before 2026.4.20, the guard on the agent-facing `gateway config.patch` and `config.apply` calls missed many operator-trusted settings. A prompt-injected model with the owner-only gateway tool could persist changes to sandbox policy, plugins, gateway auth and TLS, hook routing, MCP servers, SSRF policy and filesystem hardening. Upgrade to openclaw 2026.4.20.

Published Apr 25, 2026 · Analysis Oct 29, 2026

What an attacker can do

OpenClaw's gateway tool lets the agent change its own configuration through config.patch and config.apply. A guard is meant to stop the model from touching settings that only the operator should control. In the affected versions, that guard covered only some paths. It did not cover sandbox policy, plugin enablement, gateway auth and TLS, hook routing, MCP server configuration, SSRF policy or filesystem hardening.

The attack path is prompt injection. Text in a web page, document or message steers the model into calling the gateway tool, and the model writes a change to one of those settings. Because the change is persisted, it outlasts the session. A model could loosen the sandbox, enable a plugin, add an MCP server it controls, relax the SSRF policy or alter hook routing. Each of those widens what later, more harmful injections can do.

The advisory describes this as a model-to-operator guard bypass, not a remote unauthenticated compromise of the gateway. The model must hold the owner-only gateway tool, and the CVSS 4.0 vector shows user interaction with high integrity impact. Setups that give an agent the gateway tool while it reads untrusted content are the exposed ones.

Who is affected

The openclaw npm package before 2026.4.20, where an agent can reach the owner-only gateway tool.

npm ls openclaw

Check whether your agents have the gateway tool and what untrusted content they read.

openclaw

Ecosystem
npm
Affected
< 2026.4.20
Fixed in
2026.4.20
Check
npm ls openclaw

How to fix it

Upgrade to openclaw 2026.4.20:

npm install [email protected]

The fix commit is fe30b31a97a917ecc6e92f6c85378b6b20352422. OpenClaw now blocks model-driven gateway config changes across the wider set of operator-trusted paths, and covers per-agent overrides and patching of array entries. The advisory lists no workaround. After upgrading, check the current configuration against your known-good copy, since earlier changes persist.

How to tell if you were hit

Compare the live OpenClaw configuration with your known-good copy. Look for changes to sandbox policy, enabled plugins, gateway auth or TLS, hook routing, MCP server entries, SSRF policy and filesystem hardening. Review agent logs for config.patch and config.apply calls made by the model, and for sessions that read untrusted content just before them. The advisory names no log signature.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

Affected Packages / Versions

  • Package: openclaw (npm)
  • Affected versions: < 2026.4.20
  • Patched version: 2026.4.20

Impact

The agent-facing gateway config.patch / config.apply guard did not cover several operator-trusted settings, including sandbox policy, plugin enablement, gateway auth/TLS, hook routing, MCP server configuration, SSRF policy, and filesystem hardening. A prompt-injected model with access to the owner-only gateway tool could persist changes to those settings.

This is a model-to-operator guard bypass, not a remote unauthenticated gateway compromise. Severity is medium.

Fix

OpenClaw now blocks model-driven gateway config mutations for the broader operator-trusted path set and covers per-agent overrides and array-entry patching.

Fix commit:

  • fe30b31a97a917ecc6e92f6c85378b6b20352422

Release

Fixed in OpenClaw 2026.4.20.

Credited to zsxsoft, KeenSecurityLab, qclawer.

Questions

Is GHSA-7jm2-g593-4qrc a remote unauthenticated gateway compromise?

No. The advisory calls it a model-to-operator guard bypass. A prompt-injected model that has the owner-only gateway tool is needed.

Which settings were exposed?

Sandbox policy, plugin enablement, gateway auth and TLS, hook routing, MCP server configuration, SSRF policy and filesystem hardening.

Do changes made before the upgrade persist?

The advisory says changes could be persisted. Review the configuration after upgrading rather than assuming it is clean.

Does it have a CVE id?

The advisory data lists none. It is tracked as GHSA-7jm2-g593-4qrc.

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