CVE-2026-53815: OpenClaw: Message read actions could skip channel allowlist checks

OpenClaw checks a channel allowlist when it delivers messages. The read action skipped the same check. A caller with access to the action could request messages from any channel, including ones the allowlist is meant to keep away from that caller.

The flaw leaks data only. The CVSS vector rates confidentiality impact as high and integrity and availability as none. The attacker needs low privileges and network access to the Gateway, with no interaction. What they read depends on what flows through your channels: chat history with colleagues, pasted credentials, or the output of agent runs posted back into a channel.

The sources do not name the specific read action or the API route. The advisory ties the impact to operator configuration and to whether lower-trust input can reach the action. Gateways shared between users who do not fully trust each other face the most risk.

Affected: npm package openclaw 2026.5.18 and earlier, and 2026.5.19-beta.2 and earlier, where a message read action is enabled and reachable by a caller who is not a trusted operator.

npm ls openclaw

Check who can invoke message read actions on your Gateway. If only trusted operators can, exposure is small.

Upgrade to openclaw 2026.5.19 or later, the first stable patched version.

npm install -g [email protected]

As a stopgap the advisory says to limit message read actions to trusted operators and keep channel allowlists narrow. It also advises against sharing one Gateway between mutually untrusted users. The sources do not describe the patch.

Look in Gateway logs for message read requests that name channels outside the requesting caller's allowlist. Compare which callers invoked read actions with the list of trusted operators. The advisory gives no log strings, so the review depends on your own audit records.

Summary

Message read actions could skip channel allowlist checks. In affected versions, a lower-trust caller with access to the affected message read action could request messages without the same channel allowlist check used by normal delivery.

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 messages from a channel that was not intended for that caller. 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 2026.5.19.

Mitigations

limit message read actions to trusted operators and keep channel allowlists narrow. 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.