OpenClaw versions 2026.4.10 through 2026.7.0 allow an authenticated user who is not the owner of an external channel to toggle the Gateway's Memory Core dreaming state by issuing "/dreaming on" or "/dreaming off" commands. Because the code omits the owner permission check, non‑owners can enable or disable background memory processing or re‑enable durable memory promotion where the owner expected it to remain disabled. The practical consequences for confidentiality, integrity, or availability depend on the conversation data stored in memory and how that data is subsequently used. This flaw is a classic Missing Authorization weakness (CWE‑862).

Any OpenClaw installation deployed through npm that runs a version between and including 2026.4.10 and 2026.6.9 is vulnerable. The issue is resolved in version 2026.7.1. The affected component is the Gateway Agent that parses and processes external‑channel commands.

The vulnerability carries a CVSS score of 5.3, reflecting moderate severity. No EPSS score is reported, and the flaw has not been catalogued in CISA's KEV list. The attack requires the attacker to have legitimate channel access; the flaw does not allow arbitrary code execution or direct privilege escalation. Nevertheless, an authenticated non‑owner who can send commands can alter memory processing behavior, potentially exposing or corrupting sensitive conversation material. Organizations that rely on the Gateway’s memory management should treat this as a risk that warrants timely remediation.