When your AI agent is exposed to the internet
Self-hosted AI agents are often exposed to the public internet with minimal protection, making them targets for automated scanning and attacks.
Insecure network exposure occurs when your AI agent is accessible from the internet without proper security controls. This is common in self-hosted deployments where convenience often trumps security.
Common scenarios include: - Running on a public IP without firewall rules - Port forwarding from home routers - Exposed Docker ports with no authentication - VPN or tunnel solutions with weak security - Development servers accidentally left public
Attackers continuously scan the internet for exposed services. AI agent endpoints are particularly valuable targets because they often have access to LLM APIs, databases, and other sensitive resources.
Automated tools scan IP ranges looking for open ports and known service signatures.
Once a port is found, attackers identify what's running to find known vulnerabilities.
Search engines that index exposed services make finding targets trivial.
Trying common usernames/passwords against exposed services.
Automated tools try known exploits against identified services.
Exposed services become targets for denial of service attacks.
Security researchers demonstrated the speed of internet scanning:
1. They deployed a honeypot server with a fake AI agent endpoint 2. Within 15 minutes, the server was scanned from multiple IPs 3. Within an hour, automated credential stuffing attacks began 4. By end of day, there were hundreds of exploit attempts 5. Common attacks included prompt injection, path traversal, and known CVEs
Any exposed AI agent faces this constant barrage from day one. Without proper hardening, compromise is a matter of when, not if.
When you self-host your OpenClaw, you're responsible for addressing these risks:
Clawctl includes built-in protection against network exposure:
Your agent is never directly exposed to the internet. All access goes through our hardened gateway.
Enterprise-grade DDoS mitigation protects your agent's availability.
WAF rules block common attacks before they reach your agent.
Optional IP restrictions to limit access to known locations.
All connections are encrypted with modern TLS. No plaintext traffic.
Whether you use Clawctl or not, follow these best practices:
Clawctl includes enterprise-grade protection against this threat and many others. Deploy your OpenClaw securely in 60 seconds.