Summary
OpenClaw versions prior to 2026.2.21 start an x11vnc service for noVNC observer sessions without enforcing any authentication.
This can allow remote users with access to the host loopback interface to connect to the exposed VNC port and passively observe or actively control the sandboxed browser without credentials.
Affected context
The vulnerability occurs in OpenClaw deployments that use the sandbox browser entrypoint and rely on the noVNC observer feature.
In affected versions, x11vnc is spawned without requiring credentials, and the resulting noVNC endpoint is exposed on the host loopback interface.
Systems where loopback access is available to untrusted users or processes (for example, multi-tenant hosts, shared development servers, or environments with port-forwarding/tunneling) are at particular risk.
Exploit preconditions
- Attack surface: noVNC HTTP/WebSocket endpoint bound to the host loopback interface (e.g., 127.0.0.1:port).
- Required access: Network-level ability to reach the loopback-bound port (via local access, SSH port forwarding, proxies, or misconfigured network namespaces).
- Trigger point: Attacker initiates a browser or VNC client connection to the exposed noVNC URL/port while the sandbox browser session is running.
- Why exploitation is possible: x11vnc is configured without authentication, and no additional authorization checks are enforced before granting access to the live VNC session.
Why this happens
This issue stems from a missing authentication control on a sensitive service endpoint, mapping directly to CWE-306: Missing Authentication for Critical Function.
The sandbox browser entrypoint configures x11vnc and the associated noVNC interface as if they were inherently trusted, relying solely on loopback binding for protection.
Without explicit credential checks, any entity capable of reaching the loopback port is treated as authorized, exposing the visual and interactive browser session to unauthorized observers or controllers.
Potential impact
- Unauthorized observation of the sandbox browser session, including potentially sensitive on-screen data.
- Unauthorized interaction with the browser, such as clicking links, entering data, or changing settings.
- Hijacking of active web sessions in the sandbox (e.g., manipulating authenticated workflows or transactions visible in that browser).
- Use of the sandbox browser as a pivot point to access internal web applications or resources reachable only from that environment.
General mitigation guidance
Effective mitigation centers on enforcing explicit authentication and hardening remote access to the sandbox UI:
- Require authentication for x11vnc/noVNC access using passwords, tokens, or integration with existing identity providers.
- Gate the observer interface behind an authenticated reverse proxy or dedicated access control component, rather than relying only on loopback binding.
- Limit exposure of the noVNC port to trusted administrative users or management networks, and avoid automatic exposure in multi-tenant environments.
- Apply least-privilege principles to sandbox sessions, ensuring that even if observed or controlled, they offer minimal access to sensitive systems and data.
- Consider disabling observer functionality by default, enabling it only when explicitly needed and under controlled conditions.
How teams detect issues like this
Organizations can detect this class of issue using a combination of configuration review and network monitoring:
- Review service startup scripts, container entrypoints, and orchestration manifests for x11vnc/noVNC processes launched without authentication flags or access controls.
- Use network scanning and host-based tools to identify listening ports bound to loopback or localhost that correspond to VNC/noVNC services.
- Implement security policies or static analysis checks that flag UI-access services started without authentication or TLS on production systems.
- Monitor logs for unexpected connections to noVNC endpoints, especially from non-administrative users or automated tunnels, to identify misuse of the observer interface.

