Interactive Companion · AI Security 2026

Securing the Agentic Stack

MCP, Autonomous Agents, and the AI Pipeline

Agentic AI and the Model Context Protocol went from novelty to default in a single year. Explore the new attack surface — then break it, and defend it, hands-on.

Defense / verified
Caution
Threat / critical

Three ways in

Pick a thread and pull

Learn

A browsable repository of everything from the talk: the lethal trifecta, MCP's nine vulnerability classes, the OWASP Agentic Top 10, and the five defense layers.

Open

Lethal Trifecta

Toggle autonomy, tool access, and untrusted input. See why all three together make an agent exploitable — and how removing any one breaks the chain.

Open

Attack Sandbox

Run a simulated agent against real attack classes — tool poisoning, prompt injection, and more — then switch on defenses and watch them get blocked. Nothing real executes; it's a simulation.

Open