GHSA-grc3-2j34-p6gm: OpenClaw: message.action forwarding could send Gateway credentials to model-supplied loopback URLs

When message.action forwards an action, it can target a Gateway URL given in the action metadata. That metadata is controlled by the model. In affected versions, if the model supplied a loopback URL, OpenClaw sent the action payload with the Gateway credentials to it. The advisory does not name the token type or the function.

A prompt injection can steer the model's output, so the real attacker is anyone who can get text in front of the agent: a web page, a document or a chat message. Pairing that with a process listening on a local port, they capture the Gateway token and the payload. The CVSS 4.0 vector is local, low privileges, no user interaction, with high confidentiality impact. With the token, an attacker may act as a Gateway client, depending on what that credential allows in your setup.

The advisory leaves OpenClaw's trusted-operator model unchanged. Exposure depends on message action forwarding being enabled and on whether lower-trust input reaches the model.

Affected: npm openclaw up to and including 2026.4.29. You are exposed if message action forwarding is enabled and reachable.

npm ls -g openclaw

Drop -g for a project install. Check your configuration for message action forwarding.

Update to openclaw 2026.5.2, the first stable patched version.

npm install -g [email protected]

Until you can update, restrict message action forwarding and avoid model-supplied loopback targets. The advisory also suggests narrow channel and tool allowlists, not sharing one Gateway between mutually untrusted users, and disabling the feature if you do not need it. If a listener you do not trust ran on the host, rotate the Gateway token.

The advisory gives no log signature. Review message action calls whose metadata names a loopback address such as 127.0.0.1 or localhost. Check which local processes listen on the ports those URLs name. Look for Gateway token use from unexpected clients after such calls. Beam records agent shell commands and MCP tool calls and flags risky ones, which can help find the prompt that led to the action.

Summary

message.action forwarding could send Gateway credentials to model-supplied loopback URLs. In affected versions, model-controlled action metadata that selects a loopback Gateway URL could forward the action payload with Gateway credentials to the supplied loopback URL.

This advisory is scoped to the named feature and configuration. It does not change OpenClaw's trusted-operator model: authenticated Gateway operators, installed plugins, and intentional local execution surfaces remain trusted unless a separate policy, approval, allowlist, sandbox, or auth boundary is crossed.

Impact

When the affected feature is enabled and reachable, this could expose the token and action payload to a local listener chosen through the affected path. Practical impact depends on the operator's configuration and whether lower-trust input can reach that path.

Patched Versions

The first stable patched version is 2026.5.2.

Mitigations

restrict message action forwarding and avoid model-supplied loopback targets until patched. As general hardening, keep channel and tool allowlists narrow, avoid sharing one Gateway between mutually untrusted users, and disable the affected feature when it is not needed.