AI Fundamentals Overview

Before you can reason about why an AI system is vulnerable, you need to know how it actually works. You can't meaningfully evaluate prompt injection without understanding that there's no architectural separation between instructions and data in a context window. You can't evaluate excessive agency without understanding what a harness/loop actually controls versus what the model itself decides. You can't evaluate RAG poisoning without understanding how retrieval and chunking actually work. This section is that layer - the architecture and engineering disciplines underneath AI Security, built to be read first if you're new, or skipped straight past if you already know this material.

Why Fundamentals Before Security

Security findings are architecture-specific. "Vector and Embedding Weaknesses" (OWASP LLM09) only makes sense once you know what an embedding is and how a vector store does similarity search.
Agentic risk is a function of harness design, not just model behavior. ASI02 Tool Misuse and ASI08 Cascading Failures are consequences of how a harness wires tool calls into a loop - you have to understand the loop to understand the failure mode.
Engineering terminology keeps shifting, and so does the attack surface. Prompt engineering, context engineering, and intent engineering are successive, broader framings of the same underlying problem - each one expands what's "in scope" for both the engineer and the attacker.

The Learning Path

flowchart TD
    subgraph CA["Core Architecture"]
        A[LLM Architecture]
        B[RAG Architecture]
        C[Multi-Modal AI]
        D[Open-Weight vs Closed Models]
    end
    subgraph AS["Agentic Systems"]
        E[Agentic AI Fundamentals]
        F[Multi-Agent Systems]
        G[Harness & Loop Engineering]
    end
    subgraph ED["Engineering Disciplines"]
        H[Prompt Engineering]
        I[Context Engineering]
        J[Intent Engineering]
    end

    A --> B
    A --> C
    A --> D
    B --> E
    E --> F
    E --> G
    H --> I
    I --> J
    G --> J

    click A "../llm-architecture/" "LLM Architecture"
    click B "../rag-architecture/" "RAG Architecture"
    click C "../multi-modal-ai/" "Multi-Modal AI"
    click D "../open-weight-vs-closed-models/" "Open-Weight vs Closed Models"
    click E "../agentic-ai-fundamentals/" "Agentic AI Fundamentals"
    click F "../multi-agent-systems/" "Multi-Agent Systems"
    click G "../harness-and-loop-engineering/" "Harness & Loop Engineering"
    click H "../prompt-engineering/" "Prompt Engineering"
    click I "../context-engineering/" "Context Engineering"
    click J "../intent-engineering/" "Intent Engineering"

Section Contents

Area	Page	Focus
Core Architecture	LLM Architecture	Transformers, attention, context windows, inference mechanics - the engine everything else sits on
Core Architecture	RAG Architecture	Retrieval, chunking, embeddings, vector stores, hybrid search
Core Architecture	Multi-Modal AI	Vision/audio/video-capable models, cross-modal fusion
Core Architecture	Open-Weight vs. Closed Models	Llama/Mistral/DeepSeek vs. GPT/Claude/Gemini - licensing, deployment, and security tradeoffs
Agentic Systems	Agentic AI Fundamentals	Planning loops, tool use, ReAct and plan-execute patterns
Agentic Systems	Multi-Agent Systems	Orchestration patterns - supervisor, swarm, pipeline - and inter-agent communication
Agentic Systems	Harness & Loop Engineering	The scaffolding around a model - tool-calling loops, termination conditions, budget/turn limits
Engineering Disciplines	Prompt Engineering	Crafting input deliberately for desired model behavior
Engineering Disciplines	Context Engineering	Managing everything in the context window, not just the prompt
Engineering Disciplines	Intent Engineering	Translating ambiguous human goals into structured agent objectives (emerging term)
Reference	AI Fundamentals Resources	Papers, books, courses, and tools to go deeper on any topic above

From Fundamentals to Security

Each fundamentals page ends with a Security Angle section pointing to the matching page in AI Security - read fundamentals first if you're new to how these systems work, or jump straight to security if you already know the architecture and just need the attack surface.

Credits/References

OWASP GenAI Security Project - the security taxonomy this fundamentals layer feeds into
Anthropic: Effective Context Engineering for AI Agents
Vaswani et al., "Attention Is All You Need"