Current, content updated Oct 2026
LLM and Agentic AI Security
Udacity with Kevin Carter
Learners first attack LLM and agent systems (jailbreaks, direct, indirect and multimodal prompt injection) and then build defences such as prompt hardening, RAG access controls, secure multi-agent design, logging and approval workflows. A capstone has them red-team and harden a RAG-based research agent.
Our take
Part of Udacity's subscription; project-based and more technical than most courses here. A substantial course at about 23 hours. Strongest on owasp LLM top 10 and prompt injection. Last updated October 2026.
Best for: Security and IT professionals who need to defend against AI-driven threats or secure AI systems.
Based on the course page, syllabus and pricing, not on completing the course. How we rank
What you'll learn
- Owasp LLM top 10
- Prompt injection
- Jailbreaking
- System prompt hardening
- RAG security
- Multi-agent security
- Human-in-the-loop approval
- AI red teaming
Before you start
- Cybersecurity awareness, prompt engineering, agent tool use, LLM awareness, application architecture; fluent English