Skip to main content
Home AI Compass

AI Compass

Key ideas of AI, explained — with what can go wrong and how to secure it.

28 concepts · 5 paths · Updated October 2026

Every concept gets one card, in plain language: what it is, an example, what can go wrong and how to secure it — plus what the EU rules say, where it matters. Follow a path from the start, or jump to the term you need.

Basics

How language models work and where their limits are.

Start the path
  1. 1 LLM
  2. 2 Token
  3. 3 Prompt
  4. 4 Context window
  5. 5 Hallucination
  6. 6 Embeddings
  7. 7 Multimodal

Builder

Connecting models to your data — and testing them.

Start the path
  1. 8 RAG
  2. 9 Vector database
  3. 10 Function calling
  4. 11 Evals
  5. 12 Fine-tuning
  6. 13 Open weights

Agents

AI that plans, uses tools and acts.

Start the path
  1. 14 AI agent
  2. 15 Multi-agent
  3. 16 Model routing
  4. 17 MCP
  5. 18 Sandbox
  6. 19 Human in the loop

Security

How AI systems get attacked — and defended.

Start the path
  1. 20 Prompt injection
  2. 21 Data leakage
  3. 22 Guardrails
  4. 23 Red teaming
  5. 24 Shadow AI

EU rules

What the EU AI Act and GDPR mean in practice.

Start the path
  1. 25 AI Act risk levels
  2. 26 AI literacy
  3. 27 Transparency
  4. 28 Data residency

Why a compass?

The EU AI Act asks companies that use AI to take measures that help their people understand it (Article 4, AI literacy). Most glossaries explain what AI can do. We also explain how it fails — that's where the risk, and our daily work, is.

Share it with your team. Plain-language summaries, not legal advice.

Using AI in your company?

We build, secure and run AI systems for European companies — and help teams use them safely.

Talk to us AI Security