100 AI terms.
Without the technical hassle.

An AI system that can perform several steps to achieve a goal, such as finding information, creating a file and forwarding a task.

Artificial intelligence. Computers performing tasks that normally require human thinking.

The European law that sets rules for AI. Requirements depend on the risk and your organisation's role.

A tool that answers questions, creates content or helps with tasks. ChatGPT and Claude are well known examples.

The roles, agreements and controls used to manage AI and keep responsibility clear.

The knowledge and skills needed to use AI consciously, critically and safely.

The computational system behind an AI application. It processes input and produces an answer or prediction.

An overview of AI tools and processes, including their owner, purpose, data, location, risks and costs.

A plan that connects business goals, people, processes and technology to introduce AI safely and usefully.

A defined series of rules or steps a computer follows to solve a problem or calculate an outcome.

Changing personal data so a person can no longer reasonably be identified. This is harder than merely removing a name.

A technical interface through which two systems exchange information.

A history of actions and decisions that shows who did what, when it happened and what followed.

Letting systems perform work according to agreed steps. AI can make this more flexible but also requires additional control.

A skewed outcome caused by data, design choices or use that may unfairly disadvantage people.

A system whose reasoning is difficult to explain, which can be problematic for important decisions.

Software that performs tasks automatically. A bot does not always use AI, although many modern bots do.

A system people interact with through text or speech, using fixed answers or an AI model.

An AI assistant and model family from Anthropic for text, analysis, documents and software development.

Computing and storage accessed through the internet. Provider terms and location affect where data is processed.

An OpenAI AI agent that helps build, understand, review and modify software.

AI that can understand images or video, such as recognising objects or reading documents.

The amount of information an AI model can consider at one time.

An AI tool that works alongside a person and makes suggestions while the person remains responsible.

Personal data being lost or reaching unauthorised people, including through incorrectly shared AI input.

Grouping data by sensitivity to determine which people and AI tools may use it.

The organisation that decides why and how personal data is processed. This responsibility remains when an AI supplier is used.

A large store for different kinds of raw data that needs clear rules to remain manageable.

The origin and journey of data, including how it changes and where it is used.

Using only the data genuinely needed for a purpose. Less data usually means less risk.

The agreed region where data is stored or processed.

A collection of information used to train, test or supply a model.

Machine learning using large neural networks, widely used for language, images and speech.

Convincing AI generated or altered image, audio or video that can misrepresent what someone did or said.

Data Processing Agreement. It records how a supplier handles personal data.

A privacy risk assessment performed before processing begins, sometimes required for high risk AI.

A numeric representation of meaning that enables semantic search across text, images or other information.

Making data unreadable without the correct key, both during transfer and often during storage.

A specific internet address where software can call a service or AI model.

AI whose operation or outcome can be explained in an understandable way. Also called XAI.

Giving an AI model a few examples of the desired input and output so it can recognise the pattern.

Further training an existing AI model for a specific topic or behaviour.

A large base model that can perform many tasks and can be adapted further.

AI that creates new content such as text, images, audio, video or code.

How decisions, responsibilities, rules and controls are organised.

A technical or organisational boundary designed to prevent unwanted AI behaviour.

A convincing AI answer that is factually wrong or invented and therefore needs verification.

A person checks or decides at an important point while AI provides support.

The moment a trained AI model processes new input and creates an answer or prediction.

Information supplied to a system, such as a prompt, document, image or dataset.

An international standard for systematically managing information security.

An international management system standard for responsible development and use of AI.

An attempt to bypass an AI model's safety rules and cause unwanted output or misuse.

A managed collection of information that people or an AI application can search.

A large model, often shortened to LLM, that predicts language and can understand and produce text.

The time between a request and a system's response. Lower latency means a faster reaction.

Giving each person or tool only the access genuinely required.

Recording technical events such as requests, errors and changes for management and investigation.

AI in which a system learns patterns from examples instead of following only fixed rules.

Model Context Protocol, a standard that lets AI systems use tools and information sources in a controlled way.

A document describing an AI model's purpose, capabilities, limits, tests and risks.

The possibility that a model fails, is misused or no longer reflects reality.

An AI model that can process several types of information, such as text, images and sound.

Software for connecting systems and tasks into workflows, hosted in the cloud or on your own server.

Technology that enables computers to process human language, commonly shortened to NLP.

A computational model with connected layers that learns patterns from data.

Optical Character Recognition, technology that extracts text from scans, photos and documents.

Software whose source code is available under a licence. It is not automatically free or secure.

The result returned by a system, such as an answer, summary, prediction or file.

Information relating directly or indirectly to an identifiable person.

An extension that adds functions or connections to software.

AI that uses previous data to forecast an outcome, such as demand, failure or maintenance.

The instruction or question given to generative AI, often including context, rules and examples.

Designing, testing and improving prompts to produce useful and consistent answers.

An attack in which hidden or misleading text tries to override an AI system's instructions.

Replacing identifiable data with a code. It remains personal data because the person can still be recovered with extra information.

Agreements and tests used to monitor the quality of an AI process, often shortened to QA.

Retrieval Augmented Generation. AI first finds information in selected sources and uses it to answer.

Deliberately trying to make a system fail or be misused so weaknesses can be found early.

Developing and using AI with attention to safety, fairness, privacy, transparency and human responsibility.

Rules for how long data, prompts, answers and logs are kept and when they are deleted.

Classifying an AI use by its potential impact to determine the required controls and documentation.

Software as a Service, software purchased as an online service, usually through a subscription.

An isolated test environment where software or AI tasks can run without harming live systems.

AI used without central visibility, approval or clear agreements, leaving data, costs and ownership unclear.

Using one company login for multiple systems to improve access management.

Artificial data that imitates characteristics of real data and may reduce privacy risk when properly tested.

A core instruction that defines an AI model's role, behaviour and limits before a user asks a question.

A small unit of text processed by a language model. AI usage and costs are often measured in tokens.

An AI model triggering a controlled external function, such as reading a calendar or querying a database.

Examples from which an AI model learns. Their quality and origin affect its results.

Clearly stating that AI is used, for what purpose, with which limits and under whose responsibility.

A concrete situation where AI can create value, including the user, problem, process and desired outcome.

A database that stores embeddings and retrieves information by meaning, often used for RAG.

Strong dependence on one supplier that makes switching difficult or expensive.

A platform for building, publishing and managing websites and applications across regions.

A defined sequence of steps between people and systems. Clear workflows are the basis of good AI automation.

Explainable Artificial Intelligence, techniques that make AI outcomes and their main reasons easier to understand.

An agreement under which a supplier does not retain input and output, subject to documented exceptions.

A security principle where no access is automatically trusted and every request is verified.