Executive brief
OpenClaw is an LLM agent framework used to automate tasks and workflows. An LLM agent running within OpenClaw could bypass approval controls designed to prevent unauthorized command execution by silently modifying configuration settings. This allows a compromised or malicious agent to execute restricted commands without triggering required approval workflows, compromising operational security and auditability.
Technical details
This is an improper authorization (CWE-285) vulnerability in OpenClaw's gateway configuration handling. An LLM agent can invoke the `config.patch` method to modify critical gateway settings that control execution approval policies, bypassing the approval mechanism intended to gate sensitive operations. The vulnerability requires the agent to have access to the gateway tool module, which is typical in agent deployments. The fix, shipped in v2026.3.28, blocks protected gateway config writes by preventing agents from modifying approval-critical configuration sections. All versions up to v2026.3.24 are affected.
Affected products
- OpenClaw OpenClaw <=2026.3.24
Timeline
- 2026-03-31: disclosed
- 2026-03-27: patched: Fix committed; shipped in v2026.3.28