Guided journeys
Learning Paths
Learning Paths organize existing Learn, Attack, Defense, Labs, Research and Incident pages into guided sequences. They do not replace the canonical reference pages; every step links to the page itself.
Five paths, five goals
- 8 required steps - About 1.5 to 2 hoursFoundations
The mental model for everything else on the site
Audience: Newcomers, developers and security practitioners new to LLM applications
Classify a scenario as direct or indirect prompt injection
- 12 required steps - About 2 to 3 hoursIndirect Prompt Injection
How untrusted external content crosses context boundaries
Audience: Developers integrating external content, security reviewers, red teamers
Trace an external source into model context
- 9 required steps - About 1.5 to 2.5 hoursRAG
Retrieval mechanics, relevance vs trust, and knowledge poisoning
Audience: Engineers building retrieval systems, architects, security reviewers
Explain why relevance, rank and citation are not trust
- 15 required steps - About 2.5 to 3.5 hoursAI Agents
From model proposal to authorized execution
Audience: Agent developers, platform engineers, security architects
Separate proposal, authorization, execution and result
- 19 required steps - About 3 to 4 hoursDefense in Depth
Choosing controls by trust boundary, not by a single guardrail
Audience: Security architects, AppSec teams, technical leads
Map each control to the decision it owns
How to use these paths
- New to the topic? Follow a path in order; Foundations is the natural starting point.
- Experienced readers can jump to any step or path.
- Go-deeper items are optional and not needed to finish a path.
- Labs are controlled, client-side educational simulations, not live exploitation.
- There is no account, score or saved progress. Paths are a reading order, not a course platform.