The vulnerability arises from a logic error in OpenClaw where Signal approval reactions are incorrectly bound. When an approver reacts to a message that is not the target approval request, the reaction can be interpreted as approving or denying a pending host action if an unrelated outbound message and a pending approval exist in the same conversation. This flaw does not affect the authority of correctly identified approvers but can result in unintended execution of host‑side actions based on misinterpreted reactions.

The issue affects the npm package OpenClaw (OpenClaw:OpenClaw) in all releases prior to version 2026.7.1. Users running any of those earlier major or minor releases are subject to the risk of misbound approval events.

The CVSS score of 7.5 indicates a high severity, and the EPSS score is currently unavailable, implying no publicly known exploit activity has been reported yet. The vulnerability is not listed in CISA's KEV catalog, suggesting it is not in active exploitation. Attackers would most likely exploit this flaw by crafting messages and reactions within the same conversation thread to trigger an unintended approval or denial, though the exact attack vector is inferred from the description and not explicitly documented in the advisory.