Skill
Agent safety
Agents take actions. Treat them like untrusted users of your systems.
Do these first
What you will be able to do
- Deny network or filesystem by default
- Human-in-the-loop for irreversible actions
- Assume prompt injection from any retrieved text
Curated resources
Start with the pick, then branch out
Scores are LearnPath editorial opinions (clarity, cost, freshness, project density) — not an objective ranking. Every YouTube item is embedded officially after a click-to-load facade.
Official docs
LearnPath pickdocumentationintermediate
OWASP Top 10 for LLM Applications
OWASP50 min
Prompt injection is a security bug, not a vibe.
Score 91
Open
7h 49mAdvancedMore curatedyoutube courseadvanced
The Complete AI Security Course In 8 Hours-AI Guardrails, LLM Evals & Memory And AgentOps
Krish Naik7h 49m
An eight-hour 2026 course on securing AI systems — prompt injection, guardrails, evaluation and memory risks — the operational counterpart to AI safety theory, aimed at people deploying agents.
Score 83
Watch
8h 34mLearnPath pickMore curatedyoutube courseintermediate
AI Safety – Full Course by Safe.AI Founder on Machine Learning & Ethics (Center for AI Safety)
freeCodeCamp.org8h 34m
Eight and a half hours from a Safe.AI founder covering alignment, robustness, bias, privacy and governance with practical exercises. The most substantive free treatment of AI safety rather than a checklist of principles.
Score 81
Watch