Prompt-Injection.net
ENDE

Conceptual foundations

Learn Prompt Injection Security

Learn is the foundation layer of Prompt-Injection.net: the mental models and terminology you need before the Attack, Defense and Labs areas make full sense.

01

What this area teaches

Learn covers instructions and model context, trust boundaries, AI agents and tool calling, RAG, and the difference between model behavior and application security impact. Articles explain concepts with short synthetic examples and then link to the deeper Attack and Defense references.

Learning path

Curriculum

All nine topics are published and ordered from first principles to architecture. Model Context and Trust Boundaries intentionally point to one shared article because the concepts are inseparable in application design.

  1. Step 01 - PublishedWhat Is Prompt Injection?

    Definition, direct vs indirect, and the difference between instruction-control failure and security impact.

  2. Step 02 - PublishedInstructions and Instruction Hierarchies

    How applications and model providers try to rank instructions, and why ranking is behavior, not enforcement.

  3. Step 03 - PublishedModel Context

    What becomes model-visible and how context is assembled. Covered in Trust Boundaries and Model Context.

  4. Step 04 - PublishedTrust Boundaries

    Source vs representation vs trust, and why controls sit outside the model. Covered in Trust Boundaries and Model Context.

  5. Step 05 - PublishedPrompt Injection vs Jailbreaking

    Two overlapping terms with different emphasis, and why usage varies across sources.

  6. Step 06 - PublishedSystem Prompts

    What system prompts can and cannot do, and why they are neither secrets nor control boundaries.

  7. Step 07 - PublishedAI Agents and Tool Calling

    How model proposals turn into actions, and where authorization must decide.

  8. Step 08 - PublishedRAG and Retrieval

    How retrieved content enters context and why relevance is not trust.

  9. Step 09 - PublishedWhy Prompt Injection Is Not Just Input Validation

    Why classic input filtering alone cannot separate instructions from data in natural language.

Learning Paths

Guided journeys across the platform

The curriculum above explains individual concepts. Learning Paths instead connect Learn, Attack, Defense, Labs and evidence from Research and Incidents into ordered sequences for a specific goal.

View all learning paths

Continue in the knowledge base

After the foundations, go deeper into specific attack variants, layered defenses, or a controlled lab.

Prompt Injection Security MapSeven stages from content source to impact and evidence.Attack KnowledgeAttack variants and their technical relationships.Prompt Injection DefenseLayered controls for AI systems.Direct Prompt Injection BasicsA guided, controlled lab on model context and authorization.LabsControlled exercises without real models or systems.