Loading vulnerability…
OpenClaw agentic consent bypass via config.patch
CVE-2026-41349 is a high-severity authorization flaw in OpenClaw affecting versions prior to 2026.3.28. The vulnerability allows execution approval for LLM agents to be silently disabled through the config.patch parameter. As described in the available disclosure, a remote attacker with low privileges can abuse this behavior to bypass the product's intended consent and execution-approval security control, enabling agent actions to proceed without the expected user authorization. The issue is classified as CWE-862 (missing authorization).
What it means and what to do now: patch path, interim mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
Successful exploitation allows an attacker to bypass OpenClaw's execution approval mechanism and perform unauthorized operations without user consent. Based on the published severity vectors, the impact is high across confidentiality, integrity, and availability, implying that unauthorized agent actions may expose sensitive data, alter system or application state, and disrupt normal operations depending on the permissions and capabilities available to the affected agent or deployment.
If you can’t patch tonight, do this now.
Until patching is completed, restrict who can modify configuration inputs, especially config.patch, and limit low-privilege remote access to OpenClaw instances. Apply strict authorization checks around configuration changes and execution-approval settings, monitor for unexpected modifications to config.patch or approval-related configuration, and disable or isolate exposed functionality that permits remote configuration manipulation where operationally feasible.
Patch, then assume compromise.
Upgrade OpenClaw to version 2026.3.28 or later, as versions before 2026.3.28 are reported vulnerable. Review the vendor's fix referenced by the associated OpenClaw GitHub commit and security advisory, and ensure the execution approval mechanism can no longer be disabled through config.patch by untrusted or insufficiently authorized actors.
No public exploits tracked yet. Mallory keeps watching.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
7 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.