The AI glossary

The terms everyone throws around, explained so you actually understand them. No prior knowledge required, and no pretending everything is simple.

• What does agentic AI mean?

  Agentic AI is the umbrella term for AI that acts in multiple steps toward a goal. It’s the buzzword everyone’s selling right now, and here are the questions you should ask before buying.

• What does zero data retention mean?

  That the AI provider doesn’t keep anything you send: no questions, no answers, no training. What the promise actually covers, and what to ask your provider.

• What is a chatbot?

  The word covers everything from a website’s FAQ box to advanced AI. The difference between a chatbot, AI assistant and AI agent, without the jargon.

• What is a context window?

  The context window is the AI’s working memory: everything it can see at once. When a conversation grows longer than the window, the AI starts forgetting, unless someone has built a fix for it.

• What is a data protection impact assessment (DPIA)?

  The document your data protection officer asks for before the AI tool can be used. What it should include, when it’s required, and a ready-made template to start from.

• what is a deepfake?

  An AI-generated image, audio, or video that resembles a real person or event and can be mistaken for genuine. What the law requires, and where the line is drawn against ordinary AI images.

• What is a language model (LLM)?

  The language model is the engine behind all AI text: a program trained on huge amounts of text that predicts the next word surprisingly well. What that means in practice.

• What is a prompt?

  A prompt is what you write to an AI, and it’s the single most important thing you can improve. The same rules apply as when you ask a colleague for help.

• What is a third-country transfer?

  The legal term behind the question 'do our data end up in the US?'. What counts as a transfer, what makes it legal, and why AI tools are often involved.

• What is a token?

  AI models read and write in small chunks called tokens, and those are what you pay for. Roughly four characters per token, so longer documents cost more.

• What is AGI?

  Artificial general intelligence: AI that can do everything a human can. What the term means, why experts disagree, and why your organisation doesn’t need to wait for the answer.

• What is AIOE?

  The researchers’ way of measuring how AI-exposed a job is. How the number is calculated, why actuaries top the list and dancers are at the bottom, and why exposed doesn’t mean replaced.

• What is an AI agent?

  An AI agent gets a goal and reaches it in several steps using tools, instead of just answering a question. Here’s the difference from a chat, without the jargon.

• What is an AI hallucination?

  When AI answers confidently but wrong, it’s hallucinating: making things up. Why it happens, and which remedies actually work.

• What is an embedding?

  Embeddings are numbers that capture what a text means, so a computer can find similar content even when the words differ. The technology behind good search.

• What is an information class?

  An information class tells you how sensitive a piece of data is and where it can be processed. Four levels are enough: public, internal, confidential, and forbidden.

• What is an LLM?

  The abbreviation that pops up in every AI text: large language model. What it does, and why size matters.

• What is article 50 in the AI regulation?

  Article 50 is the EU’s transparency rule: people should know when they encounter AI, and AI-generated content must be labelled. The requirements apply from August 2026.

• What is digital sovereignty?

  The ability to control your own data and systems, even when the world around you changes. Why the question has moved from the IT department to the boardroom.

• What is fine-tuning?

  Retraining a ready-made AI model on your own material. When you need it, when you don’t, and why most organisations get by perfectly well without it.

• What is generative AI?

  AI that creates new content instead of just sorting or recognising things: text, images, audio, and code. The driving force behind the whole AI wave, explained without the jargon.

• What is RAG?

  RAG means the AI looks up your own material before answering, instead of guessing from memory. It’s the technology behind answers with sources.

• What is shadow AI?

  When employees use AI tools unknown to their employer, with private accounts and company data. Why it happens everywhere, and what actually helps.

• What is the AI regulation (AI Act)?

  The EU's law on artificial intelligence: what it requires from ordinary organisations, which parts are already in force, and why most of it is less scary than it sounds.

• What is training data?

  The material an AI model learns from, and the question every customer should ask: is the model trained on what we input? Here’s how it works.