OpenClaw Agentic Consent Bypass Vulnerability (CVE-2026-41349)
OpenClaw before version 2026.3.28 contains an agentic consent bypass vulnerability (CVE-2026-41349) that allows LLM agents to silently disable execution approval via the config.patch parameter, potentially enabling remote attackers to bypass security controls and execute unauthorized operations without user consent.
CVE search metadata
CVE search record: CVE-2026-41349. Severity: high. CVSS: 8.8. KEV: no. Product: OpenClaw. Brief: OpenClaw Agentic Consent Bypass Vulnerability (CVE-2026-41349). Brief link: https://feed.craftedsignal.io/briefs/2026-04-24-openclaw-bypass/
Triage facts
Revision 2- Applicability
- OpenClaw
- Evidence
- Unknown
- Severity
- High
- Sources
- 4 public sources
- Freshness
- Revision 2Apr 24, 12:00 UTC
- Detection
- 2 rules
- Vulnerabilities
- 1 CVE
Revision history 1 entry · Revision 2
- Revision 2
Repaired a legacy canonical identity collision
Changed: slug
OpenClaw, a framework for building LLM agents, is vulnerable to an agentic consent bypass. Specifically, versions prior to 2026.3.28 are susceptible to CVE-2026-41349, where an LLM agent can silently disable execution approval through manipulation of the config.patch parameter. This vulnerability allows remote attackers to bypass security controls that are meant to ensure user consent before executing operations. The vulnerability was reported on April 23, 2026. Successful exploitation could lead to unauthorized actions being performed by the LLM agent without the user's knowledge or permission, undermining the intended security model of OpenClaw.
Attack Chain
- An attacker gains initial access to an OpenClaw instance or application using it, potentially through compromised credentials or by exploiting another vulnerability.
- The attacker crafts a malicious
config.patchparameter designed to disable execution approval. - The attacker injects the crafted
config.patchinto the OpenClaw configuration, exploiting the vulnerability. - The OpenClaw application processes the modified configuration, effectively bypassing consent checks.
- The LLM agent, now operating without consent requirements, initiates unauthorized actions.
- These unauthorized actions can include data exfiltration, system modification, or other malicious activities depending on the LLM agent's capabilities.
- The attacker monitors the LLM agent's activities, leveraging its capabilities for further malicious purposes.
Impact
Successful exploitation of this vulnerability allows an attacker to bypass intended security controls and execute unauthorized operations. The severity of impact depends on the privileges and capabilities granted to the LLM agent. In a worst-case scenario, an attacker could gain complete control over systems accessible to the LLM agent, leading to significant data breaches, system compromise, or financial loss. While the exact number of affected organizations is unknown, any deployment of OpenClaw before version 2026.3.28 is potentially vulnerable.
Recommendation
- Upgrade OpenClaw to version 2026.3.28 or later to remediate CVE-2026-41349.
- Implement monitoring for unexpected modifications to the OpenClaw configuration, especially those involving the
config.patchparameter. Deploy the Sigma ruleDetect OpenClaw Config Patch Manipulationto detect such modifications. - Review and restrict the permissions granted to LLM agents to minimize the potential impact of unauthorized actions.
- Monitor network traffic for suspicious activity originating from OpenClaw instances. Use the IOC information provided by VulnCheck's advisory for threat hunting.
Detection coverage 2
Detect OpenClaw Config Patch Manipulation
highDetects modifications to OpenClaw configuration files that contain 'config.patch', indicating a potential attempt to bypass security controls.
Detect OpenClaw Unauthorized Execution
mediumDetects execution of critical OpenClaw functions without prior authorization, potentially indicating a bypass of consent mechanisms.
Detection queries are available on the platform. Get full rules →