Lab

Securing OpenClaw with Tetragon

When your AI assistant goes rogue, eBPF saves the day.

Your personal AI assistant, OpenClaw, is running on your server. It can read your files, call APIs, and automate tasks. But what happens when it unearths that resignation letter you never sent, leaks your friend's surprise party plans, messages your friends something embarrassing, and books the wrong restaurant?

In February 2026, researchers found a critical vulnerability in OpenClaw's core gateway that allowed any website to silently hijack a developer's AI agent.

In this lab, you will witness the chaos unfold, observe the damage with Tetragon's security observability, and then enforce runtime security policies to stop OpenClaw from ever misbehaving again — all powered by eBPF in the Linux kernel.

  • 📦 The Setup

    Install Tetragon, deploy OpenClaw, and set the scene

  • 🔥 The Chaos

    Watch OpenClaw go rogue with your private data

  • 🔎 Observe the Damage

    Use Tetragon events to understand exactly what happened

  • 🔒 Lock It Down

    Write Tetragon enforcement policies to stop OpenClaw

  • 🚨 Verify Enforcement

    Re-trigger the rogue actions and watch Tetragon block them

  • 🎓 Exam: Securing the Gateway

    Write a Tetragon policy to block the OASIS exploit