fischer³ Agentic AI Security Education¶

Welcome to¶

AI agents are no longer a concept on the horizon — they are here, and they are talking to each other. The Agent-to-Agent (A2A) protocol, the Model Context Protocol (MCP), and the systems that tie them together represent a new frontier in software architecture. With that frontier comes a new set of security challenges that most existing resources barely touch.

This project exists because the security of agentic AI systems deserves the same rigorous, hands-on treatment we give to application security — but almost nobody is teaching it that way yet. You won't find a glossy overview and a "good luck" here. Instead, you'll find deliberately vulnerable code, real threat models, progressive security implementations, and the explanations that connect all the dots. Whether you're a developer building your first agent, a security professional evaluating an AI system, or an architect designing something that needs to work at production scale, this material was built with you in mind.

The core philosophy is simple: you learn security best by seeing what happens when it's missing. Every example in this project starts from an insecure baseline and evolves through increasingly robust implementations. By the time you reach the production-ready stages, you won't just know what to do — you'll understand why each control exists and what breaks without it.

What This Project Covers¶

Where to Begin¶

The right starting point depends on your background and what you're trying to accomplish.

New to Agentic AI?¶

Start here if you're unfamiliar with how AI agents communicate and collaborate, or if you want a grounded understanding before diving into security specifics.

1. A2A Overview — What agent-to-agent communication is and why it matters
2. Core Concepts — The building blocks of A2A systems
3. MCP Summary — How agents connect to tools
4. Integration Summary — How A2A and MCP fit together

Estimated time: 2–3 hours

A Developer Ready to Build¶

You understand the basics and want to build agents that are secure from the start. This path gets you to production-ready patterns efficiently.

1. A2A Overview — Quick orientation
2. Security Best Practices — The controls that matter most
3. Message Validation Patterns — Eight-layer defense in depth
4. MCP Fundamentals — Protocol mechanics for tool integration
5. Hands-On Examples — See it all in working code

Estimated time: 4–6 hours

A Security Professional Evaluating Agentic Systems¶

You need to understand the threat landscape, identify weaknesses, and know what good looks like. This path is built around exactly that workflow.

1. Threat Model — What attacks actually look like against agent systems
2. Authentication Tags — How cryptographic verification works in A2A
3. Code Walkthrough — Vulnerable vs. secure, side by side
4. Session State Security — A frequently overlooked attack surface
5. Security Checklist — 200+ item audit tool

Estimated time: 6–8 hours

Not a Technical Professional¶

You don't write code, but you need to understand what's at stake when organizations deploy AI agents — and how to ask the right questions.

1. AI Collaboration Fundamentals — No code required
2. Security for Non-Technical Audiences — Risk in plain language
3. Agent Security Article — The executive summary and business case sections

Estimated time: 2 hours

Hands-On Examples¶

Each example follows the same pattern: start with an intentionally vulnerable implementation, then progressively harden it stage by stage. By comparing stages, you see exactly what each security control buys you — and what happens without it.

MCP Examples¶

The Eight-Layer Security Framework¶

A recurring theme throughout this project is defense in depth — the idea that no single security control is sufficient on its own. The eight-layer framework provides a structured way to think about this:

The presentation materials provide extensive training content around this framework, including a 200+ item security checklist, slide decks for team training and executive briefings, and a detailed article walking through all eight layers with real-world examples.

Key Concepts at a Glance¶

Agent Card — A standardized identity declaration. Every agent in an A2A system publishes one, advertising its capabilities, supported protocols, and authentication requirements. Learn more in Agent Cards.

Agent Registry — A discovery service where agents register themselves and other agents look them up by capability. Central to how multi-agent systems find and coordinate with each other. Learn more in Registry Patterns.

MCP Server — A process that exposes tools and resources to AI agents via the Model Context Protocol. Think of it as a secure adapter between an agent and an external system. Learn more in MCP Fundamentals.

Session State Security — Managing the lifecycle and integrity of agent-to-agent sessions. This is one of the most overlooked attack surfaces in agentic systems, and one of the most thoroughly covered topics in this project. Learn more in Session State Security.

🤝 Contributing¶

Contributions are welcome and encouraged.

• Found a bug? Open an issue on GitHub
• Want to contribute content? Submit a pull request
• Have questions? Start a discussion
• Found a security issue? Report responsibly to robert@fischer3.org

Ways to help: improve documentation clarity, add new examples, report security findings, translate to other languages, or share your own implementations.

📝 License¶

This project is released under the MIT License.

📬 Contact¶

Project Maintainer: Robert Fischer
Email: robert@fischer3.org

Last Updated: January 2026
Version: 3.0
Status: Active Development