How to Safely Evaluate OpenClaw with Cloudflare’s Moltworker Sandbox
OpenClaw, an open-source AI agent, has been gaining popularity among developers, with over 21,000 publicly exposed deployments in just under a week. However, security concerns have been raised due to vulnerabilities such as CVE-2026-25253 and CVE-2026-25157, which can lead to remote code execution and command injection.
Bitdefender’s GravityZone telemetry has confirmed that employees are deploying OpenClaw on corporate machines with single-line install commands, potentially exposing sensitive data. Additionally, security audits have revealed that a significant number of skills on the ClawHub marketplace contain critical security flaws, putting credentials at risk.
Furthermore, Moltbook, a social network built on OpenClaw infrastructure, left its entire database publicly accessible, exposing API authentication tokens and email addresses. This highlights the importance of proper security measures when deploying AI agents in corporate environments.
To address these security concerns, Cloudflare has introduced the Moltworker framework, which provides a secure evaluation environment for OpenClaw. By running the agent in isolated, ephemeral containers with encrypted storage and Zero Trust authentication, security risks can be minimized.
Setting up a secure evaluation instance with Moltworker is straightforward. By configuring storage and billing, generating tokens, deploying the agent, and enabling Zero Trust authentication, security leaders can create a controlled environment for testing OpenClaw.
During the evaluation period, it is crucial to use synthetic data and test accounts to prevent exposure of sensitive information. Observing how the agent handles tasks like scheduling and web research can provide valuable insights into its capabilities without risking data breaches.
Additionally, conducting adversarial tests within the sandbox environment can help identify potential vulnerabilities and assess the agent’s behavior in response to malicious inputs. By monitoring outbound connections, credential handling, and container boundaries, security leaders can ensure that the agent operates securely.
Ultimately, establishing a secure evaluation framework for AI agents like OpenClaw is essential for mitigating security risks and preventing data breaches. By following best practices and leveraging tools like Cloudflare’s Moltworker, organizations can harness the productivity gains of agentic AI while safeguarding against potential threats.