Learning roadmap
AI Security & Red-Teaming
Threat-model your LLM features, defend against prompt injection, attack your own agent before someone else does, and build the guardrails and response plan that turn "AI breach" from catastrophe into incident.
4 stages~8 h
That leaves 4 stages, ~8 hours of focused work. You can clear it in any order, but the stages compound — try them top-to-bottom your first time.
- You'll pick up:
untrusted input surface, tool privilege boundary, assets and adversaries, data flow map, owasp llm top ten
Practice quizzesLab taskBuild a "threat model canvas": a feature diagram where you drop asset badges (user data, credentials, tools, wallets) on components, draw the data flows, and tag each flow with its trust level — red flows get a defense requirement.
Build it in the LabCheckpointYour feature has a one-page threat model a security reviewer can attack line by line.
What you'll be able to do
- Write the threat model for an LLM feature: assets, adversaries, and the flows that move data.
- Layer real defenses against direct and indirect prompt injection — and know which ones fail quietly.
- Run a scheduled red-team suite that attacks your agent with jailbreaks and exfiltration attempts.
- Respond to a security incident with a plan written before the incident.
Or try a related path