ClawdBot Security Guide

Clawdbot (now renamed Moltbot) is an open-source, self-hosted AI assistant that runs on your own hardware or server and can-do things, not just chat.

It was created by developer Peter Steinberger in late 2025.

It connects your AI model (OpenAI, Claude, local models via Ollama) to real capabilities: automate workflows, read/write files, execute tools and scripts, manage emails/calendars, and respond through messaging apps like WhatsApp, Telegram, Discord and Slack.

You interact with it like a smart assistant that actually takes action based on your input.

What is it used for?

Clawdbot functions as a "digital employee" or a "Jarvis-like" assistant that operates 24/7. Because it has direct access to your local filesystem and system tools, it can perform proactive tasks that standard AI cannot:

• Communication Hub: It lives inside messaging apps like Telegram, WhatsApp, or Slack. You text it commands, and it can reply, summarize threads, or manage your inbox.
• Proactive Automation: It can monitor your email, calendar, and GitHub repositories to fix bugs while you sleep, draft replies, or alert you to flight check-ins.
• System Execution: It can run shell commands, execute scripts, manage files, and even control web browsers to perform actions like making purchases or reservations.
• Persistent Memory: It maintains long-term context across conversations, remembering your preferences and past tasks for weeks or months.

Below is a sample deployment architecture of Clawdbot:

Security risks associated with Clawdbot

Clawdbot is a high-privilege automation control plane. Since it manages agents, tools, and multiple communication channels, it presents serious security risks.

Control plane exposure & misconfiguration

• Exposure: Misconfigured dashboards and reverse proxies have left hundreds of control interfaces open to the internet.
• Authentication Failures: Some setups treat remote connections as local, letting attackers bypass authentication.
• Data Theft: Unsecured instances can expose API keys, conversation logs, and configuration data.
• System Takeover: In certain cases, attackers can run commands on the host with elevated privileges.

Prompt injection & tool blast radius

• Manipulation: Malicious or untrusted content can trick the AI into using tools in unintended ways.
• Blast Radius: Access to high-privilege tools like shell commands or admin APIs means a prompt injection could lead to data theft or lateral movement across the network.
• Model Weakness: Older or poorly aligned AI models are more likely to ignore safety instructions, increasing risk.

Social engineering and user level abuse

• Deception: Attackers can manipulate the bot to extract personal or environment-specific information.
• Account Misuse: Connected commerce tools could be used for unauthorized purchases.
• Phishing: A compromised bot can send malicious links or scripts to contacts.
• Upstream Data Exposure: Prompts and tool outputs sent to AI providers can create privacy or compliance issues if not carefully managed.

Data privacy, logs, and long term memory

• Sensitive Data Exposure: The gateway stores conversation histories and memory, which may include personal or business information depending on usage.
• Dashboard and Host Vulnerabilities: Exposed dashboards or weak host protections can allow attackers to access past chats, file transfers, and stored credentials (API keys, tokens, OAuth secrets), turning the instance into a data exfiltration point.
• Upstream Data Risk: Prompts and tool outputs are sent to AI providers. Without proper scoping and data classification, this can create privacy and compliance issues.

Ecosystem risks: hijacked branding, fake installers, and scams

• Hijacked Accounts: After a rebrand, original social media and GitHub handles were exploited by scammers promoting fake crypto tokens.
• Malware Risk: Users searching for the tool may encounter backdoored versions or fake installers designed to compromise their systems.

Network and Remote Access Risks

• Browser Control: Tools that let the bot control a browser can expose local or internal network resources if not secured.
• Tunneling Errors: Misconfigured reverse proxies or tools like Tailscale may grant attackers unintended access to private networks.

Recommendations for securing Clawdbot

Based on the official GitHub repository, documentation, and expert audits from January 2026, here are the recommendations for securing your instance.

Lock Down the Gateway

Bind the Clawdbot gateway to loopback (127.0.0.1) and never expose it directly to the internet. If remote access is required, use private mesh solutions such as Tailscale or Cloudflare Tunnel. Always enable gateway authentication using tokens or passwords.

References:

• Official GitHub Security Overview
• Clawdbot Remote Access Documentation

Enforce Strict Access Controls

Restrict who can interact with Clawdbot by enforcing DM pairing or allowlists. Avoid wildcard policies in production. In group chats, require explicit mentions before the bot processes messages.

Reference:

• Official GitHub SECURITY.md

Isolate the Runtime Environment

Run Clawdbot on dedicated hardware or a dedicated VM/container. Avoid running it on your primary workstation. Use Docker sandboxing with minimal mounts and dropped capabilities.

References:

• Clawdbot Getting Started Guide
• Official GitHub Security Overview

Sandbox and Restrict Tools

Enable sandboxing for all high-risk tools such as exec, write, browser automation, and web access. Use tool allow/deny lists and restrict elevated tools to trusted users only.

Reference:

• Official GitHub Security Overview

Apply Least Privilege to Agent Capabilities

Disable interactive shells unless strictly necessary. Limit filesystem visibility to read-only mounts where possible. Avoid granting elevated privileges to agents handling untrusted input.

Reference:

• Official Clawdbot Documentation

Secure Credentials and Secrets

Store secrets in environment variables, not configuration files or source control. Apply strict filesystem permissions to Clawdbot directories and rotate credentials after any suspected incident.

Reference:

• Official Clawdbot Security Documentation

Continuous Auditing and Monitoring

Regularly run built-in security audit and doctor commands to detect unsafe configurations. Monitor logs and session transcripts for anomalous behavior or unexpected access.

Reference:

• Official GitHub Security CLI Documentation

Harden Browser Automation

Treat browser automation as operator-level access. Use dedicated browser profiles without password managers or sync enabled. Never expose browser control ports publicly.

Prompt-Level Safety Rules

Define explicit system rules that prevent disclosure of credentials, filesystem structure, or infrastructure details. Require confirmation for destructive actions.

Reference:

• Official Clawdbot Security Documentation

Incident Response Preparedness

Maintain a documented response plan. If compromise is suspected: stop the gateway, revoke access, rotate all secrets, review logs, and re-run security audits.

Reference:

• Official Clawdbot Security Documentation

Summary

ClawdBot is a high-privilege AI agent that can act on your system, not just chat. Its main risks come from exposed gateways, weak access controls, and powerful tools combined with prompt injection or social engineering, which can lead to system compromise and data loss. To use it safely, lock the gateway to localhost with authentication, restrict who can interact with it, isolate its runtime, minimize tool permissions, and monitor it continuously.

Disclaimer: AI tools were used to research and edit this article. Graphics are created using AI.

References:

• Your Clawdbot AI Assistant Has Shell Access and One Prompt Injection Away from Disaster
• ClawdBot: Setup Guide + How to NOT Get Hacked
• 10 ways to hack into a vibecoder's clawdbot & get entire human identity (educational purposes only)
• Hacking clawdbot and eating lobster souls
• Eating lobster souls Part II: the supply chain
• Eating lobster souls Part III (the finale): Escape the Moltrix

About the Author

Eyal Estrin is a cloud and information security architect and AWS Community Builder, with more than 25 years in the industry. He is the author of Cloud Security Handbook and Security for Cloud Native Applications.

The views expressed are his own.