OpenClaw and Langflow critical vulnerabilities enable authentication bypass
360 Security researchers discovered zero-day vulnerabilities in OpenClaw Gateway and Langflow that allow attackers to bypass authentication and gain control of AI agent systems. The OpenClaw founder confirmed the vulnerability, which was reported to China's national vulnerability database. Langflow's critical vulnerability was exploited within hours of disclosure.
What happened
360 Security Cloud Team discovered a zero-day WebSocket authentication bypass vulnerability in OpenClaw Gateway. OpenClaw founder Peter confirmed the vulnerability in an official email to 360.
The vulnerability allows attackers to bypass permission authentication through WebSocket and gain control of the AI agent gateway, potentially causing system resource exhaustion or complete system collapse.
Langflow also disclosed a critical vulnerability (CVE-2026-33017) that was exploited within 20 hours of public disclosure.
Reported impact
Attackers exploiting the OpenClaw vulnerability could achieve full control over AI agent gateway systems and cause target systems to fail.
The Langflow vulnerability was actively exploited following its public disclosure.
Index analysis
The vulnerabilities highlight security risks as AI agents transition from conversational tools to execution systems, with risks extending from model layer to interface, skill invocation chain, and system permission layers.
Common industry risks identified include exposed public interfaces, malicious skill injection, prompt injection, and lack of audit mechanisms in agent behavior.
Governance controls
360 reported the OpenClaw vulnerability to China's National Information Security Vulnerability Sharing Platform (CNVD) to enable rapid mitigation across networks.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.