OpenClaw agents prior to version 2026.7.1 allow an authorized external‑channel sender that is not the resource owner to toggle the global Active Memory setting without owner verification. This bypass permits the sender to permanently enable or disable memory recall for all future sessions or to re‑enable the recall when the owner intended it to remain disabled. The flaw does not grant arbitrary code execution, but it does undermine the confidentiality and integrity of the gateway’s memory state by permitting unintended state changes. The vulnerability is classified as a missing owner authorization check on Active Memory toggle operations (CWE‑862).

All installations of the OpenClaw agent gateway running a version prior to 2026.7.1 are affected. The product is distributed via npm under the OpenClaw namespace. Version information for the affected releases is not enumerated in the advisory, but the fix is delivered in release 2026.7.1 and later.

The CVSS score of 5.3 indicates a moderate risk level. The EPSS score is not available, and the vulnerability is not currently listed in the CISA KEV catalog. The likely attack vector is an External Channel Sender that is authenticated but not an owner; the attacker must be able to send control commands to the gateway. No exploitation code or public proof‑of‑concept is disclosed, so an attacker would need to leverage legitimate channel access and then exploit the missing owner check. Overall, the risk is moderate but can lead to unintended state changes that may impact operational integrity.