Agentic AI (LangChain and LangGraph) · Production, Evaluation and Security · Advanced · 30 min read
Guardrails and Prompt-Injection Defence for AI Agents: Threat Model, Controls and Middleware
Secure tool-using agents. Learn the threat model (direct and indirect prompt injection, data exfiltration, excessive agency), the lethal trifecta, defence in depth with least privilege, input and output guardrails, PII handling, human approval, sandboxing and allow-lists, how to implement them as LangChain middleware, and how to test your defences.
Create an account to start
Create a free account to see what this lesson covers
- Every free lesson in every section, in full
- Track your progress and pick up where you left off
- Free practice questions and an incident simulation
- Takes under a minute: name, email and a password
No card needed. Full access to Agentic AI (LangChain and LangGraph) is a paid upgrade you can choose later.