AI Security & Trust

Why This Matters

Prompt injection, poisoned retrieval, and tool chaining are not traditional app-sec bugs. They appear because the model is an untrusted planner sitting on privileged tools.

Architecture Position

Trust Layer, wrapping every other layer: input, knowledge, agent actions, identity, and runtime isolation. See it on the architecture map.

Start Here

Building Enterprise AI Agents

Production harness: runtime, orchestration, tools, memory, RAG, security, observability, and production controls.

Advanced · 40 min

Context Engineering

17 prompt and context engineering techniques for building reliable AI systems.

Intermediate · 28 min

Core Concepts

Threat Modeling

AI threat models
Agent threat models
Trust boundaries

Input Attacks

Prompt injection
Indirect prompt injection
Jailbreaking

Knowledge Attacks

RAG poisoning
Context poisoning
Memory poisoning

Agent Attacks

Tool poisoning
Goal manipulation
Excessive agency
Capability chaining

Identity Attacks

Credential theft
Confused deputy
Privilege escalation

Data Security

Data leakage
Cross-tenant exposure
Secret exposure
Exfiltration

Runtime Security

Sandboxing
Execution isolation
Capability controls

Protocol Security

MCP security
Agent-to-agent trust

Assurance

Red teaming
Security evals
Policy enforcement
Audit

Advanced

Advanced material

Deeper guides for this domain are on the way.

Coming Soon

Architecture Patterns

Applied patterns for this domain live in the pattern library.

Browse architecture patterns

Case Studies

Enterprise walkthroughs that apply this topic in a full system.

Browse case studies

Roadmaps

This topic appears in the Enterprise AI Architect path.

Follow the Enterprise AI Architect roadmap