The artificial intelligence and data terms you’ll hear most often in an AI project, explained in plain language. Where one exists, we add the reference definition from an official or primary source—the EU AI Act, the OECD, NIST, ISO/IEC standards, or the original research paper—with a link so you can check it.

Jump to: A · B · C · D · E · F · G · H · I · K · L · M · O · P · R · S · T · V
A
AI agent
A software system that, given a goal, plans and carries out tasks autonomously: it queries data, uses tools (APIs, systems, or searches), and decides on next steps within defined limits and permissions. See service: AI agents and multi-agent systems →
Reference definition — Russell and Norvig, “Artificial Intelligence: A Modern Approach,” Chapter 2, Section 2.1: “An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.” View source (opens in a new tab)
AI governance
The set of policies, roles, processes, and controls that ensure AI is used in a way that is safe, ethical, legal, and aligned with the organization’s goals. See service: AI consulting and strategy →
AI literacy
The knowledge and skills needed to use AI in an informed way, understanding its opportunities, its risks, and the harm it can cause. See service: Corporate AI training →
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(56): “skills, knowledge and understanding that allow providers, deployers and affected persons, taking into account their respective rights and obligations in the context of this Regulation, to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause” View source (opens in a new tab)
AI system
A software-based system that, from the data it receives, generates predictions, content, recommendations, or decisions with some degree of autonomy. It is the core concept on which AI laws and standards are built. See service: AI consulting and strategy →
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(1): “a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments” View source (opens in a new tab)
Reference definition — OECD, Recommendation of the Council on Artificial Intelligence (AI system definition revised in 2023): “An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.” View source (opens in a new tab)
Reference definition — UNESCO, Recommendation on the Ethics of Artificial Intelligence (2021): UNESCO does not set a single definition of AI. It approaches AI systems as systems that can process data and information in a way that resembles intelligent behavior, typically including aspects of reasoning, learning, perception, prediction, planning, or control. (paraphrase by Wasivia, not a verbatim quote) View source (opens in a new tab)
B
Business intelligence (BI)
The set of processes and tools used to collect, integrate, and visualize a company’s data in order to support decision-making. See service: Data analytics and business intelligence →
C
Chain of thought
A technique in which the model is asked to reason, or shown reasoning, step by step before giving its final answer, which tends to improve results on complex problems. See service: Corporate AI training →
Reference definition — Wei et al., “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models” (2022): “a simple method called chain of thought prompting, where a few chain of thought demonstrations are provided as exemplars in prompting” View source (opens in a new tab)
Context window
The maximum amount of text, measured in tokens, that a language model can take into account at once, including instructions, documents, and the conversation.
D
Dashboard
A visual panel that brings together the key indicators of a department or the whole company, usually updated automatically, so you can track performance at a glance. See service: Data analytics and business intelligence →
Data analytics
The process of examining, cleaning, transforming, and modeling data to obtain useful information that supports decision-making. See service: Data analytics and business intelligence →
Data poisoning
An attack that manipulates the data a model is trained on in order to alter its behavior. Read more on our blog (in Spanish) →
Reference definition — NIST, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations (NIST AI 100-2e2025), Section 2.1.1: “In a data poisoning attack, an adversary controls a subset of the training data by either inserting or modifying training samples.” View source (opens in a new tab)
Deep learning
A type of machine learning based on neural networks with many layers, especially effective with images, audio, and text. It is the foundation of today’s language models.
Deepfake
An image, audio, or video created or altered with AI that looks real and can make people believe that something happened or that someone said something they did not say. Read more on our blog (in Spanish) →
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(60): “AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful” View source (opens in a new tab)
Deployer
Under the EU AI Act, the organization or person that uses an AI system in its professional activity. A company that rolls out an AI assistant for its team usually has this role.
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(4): “a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity” View source (opens in a new tab)
E
Embedding (vector representation)
A representation of a piece of text, an image, or other data as a list of numbers (a vector) that captures its meaning, so that similar content ends up close together. See service: Generative AI, RAG and knowledge bases →
ETL (extract, transform, load)
A process that extracts data from different sources, cleans and transforms it, and loads it into a common destination, such as a data warehouse, for analysis. See service: Data analytics and business intelligence →
Evasion attack
An attack that slightly modifies an input, sometimes imperceptibly, so that the model misclassifies it. Read more on our blog (in Spanish) →
Reference definition — NIST, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations (NIST AI 100-2e2025), Section 2.2: “In an evasion attack, the adversary’s goal is to generate adversarial examples: samples whose classification can be changed to an arbitrary class of the attacker’s choice – often with only minimal perturbation.” View source (opens in a new tab)
F
Fine-tuning
The process of further training an existing model on specific data to specialize it in a task, a style, or a domain. See service: Generative AI, RAG and knowledge bases →
Foundation model
A model trained on large volumes of general data that can be adapted to many specific tasks, such as large language or image models. See service: Generative AI, RAG and knowledge bases →
Reference definition — Bommasani et al., “On the Opportunities and Risks of Foundation Models,” Stanford (2021): “We call these models foundation models to underscore their critically central yet incomplete character.” View source (opens in a new tab)
G
General-purpose AI model
A model, such as a large language model, that can perform many different tasks and can be integrated into other applications. The EU AI Act sets specific obligations for these models. See service: Generative AI, RAG and knowledge bases →
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(63): “an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market” View source (opens in a new tab)
Generative AI
A type of artificial intelligence that can create new content (text, images, audio, or code) from instructions, based on patterns learned from large volumes of data. See service: Generative AI, RAG and knowledge bases →
Guardrails
Technical controls that limit what an AI system can receive, respond with, or do; for example, content filters, validations, or restrictions on actions. See service: Generative AI, RAG and knowledge bases →
H
Hallucination
A response from a generative AI model that sounds convincing but is false or not supported by the data. It can be reduced with techniques such as RAG, precise instructions, and systematic evaluation of responses. See service: Generative AI, RAG and knowledge bases →
Reference definition — NIST, Generative AI Profile (NIST AI 600-1, 2024), Section 2.2, “Confabulation”: ““Confabulation” refers to a phenomenon in which GAI systems generate and confidently present erroneous or false content in response to prompts. […] These phenomena are colloquially also referred to as “hallucinations” or “fabrications.”” View source (opens in a new tab)
I
Inference
Using an already trained model to produce a prediction or a response from new data. It is the phase in which the model works in production. See service: Predictive models and machine learning →
ISO/IEC 42001
An international standard for implementing an AI management system in an organization: policies, roles, risks, and continual improvement, much as ISO/IEC 27001 does for information security. See service: AI consulting and strategy →
Reference definition — ISO/IEC 42001:2023 (official abstract): “This document specifies the requirements and provides guidance for establishing, implementing, maintaining and continually improving an AI (artificial intelligence) management system within the context of an organization.” View source (opens in a new tab)
K
Knowledge base
An organized repository of a company’s information (documents, procedures, FAQs) prepared so that people, AI assistants, and agents can query it. See service: Generative AI, RAG and knowledge bases →
KPI (key performance indicator)
A metric that measures progress toward a business goal, such as monthly sales, response time, or customer churn rate. See service: Data analytics and business intelligence →
L
LLM (large language model)
An AI model trained on enormous amounts of text to understand and generate language. It is the technology behind conversational assistants and many AI agents. See service: Generative AI, RAG and knowledge bases →
M
Machine learning
A branch of artificial intelligence in which systems learn patterns from data, instead of following explicitly programmed rules, to make predictions or classifications. See service: Predictive models and machine learning →
MCP (Model Context Protocol)
An open protocol, introduced by Anthropic in 2024, that standardizes how AI applications connect to external tools and data sources. See service: AI agents and multi-agent systems →
Reference definition — Model Context Protocol official documentation: “MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems.” View source (opens in a new tab)
Reference definition — Anthropic, announcement of November 25, 2024: “The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools.” View source (opens in a new tab)
MLOps
Practices for deploying, monitoring, versioning, and maintaining machine learning models reliably in production, inspired by DevOps. See service: Predictive models and machine learning →
Model evaluation (evals)
A set of systematic tests to measure the quality, accuracy, and safety of an AI model before and after putting it into production. See service: Predictive models and machine learning →
Multi-agent architecture
A design in which several AI agents work together: a coordinator agent splits the work and assigns it to specialist agents, each with its own tools and responsibilities. See service: AI agents and multi-agent systems →
O
Overfitting
A problem in which a model learns the training data too well, including its noise, and therefore performs poorly on new data. See service: Predictive models and machine learning →
P
Predictive model
A statistical or machine learning model that estimates a future value or probability, such as demand or churn risk, from historical data. See service: Predictive models and machine learning →
Prompt
The instruction or input text given to a generative AI model to get a response. How clear it is and how much context it provides directly affect the quality of the result. See service: Corporate AI training →
Prompt injection
An attack in which malicious text, hidden in a document, a web page, or a message, tries to make an AI model or agent ignore its instructions and do something unauthorized. Read more on our blog (in Spanish) →
Reference definition — NIST, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations (NIST AI 100-2e2025), Section 3.3: “A subset of these attacks, in which the main user provides in-context instructions that are appended to higher-trust instructions like those provided by the application designer (such as the model’s system prompt), are known as direct prompt injection attacks.” View source (opens in a new tab)
Reference definition — NIST, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations (NIST AI 100-2e2025), Section 3.4: “unlike in direct prompt injection attacks, indirect prompt injection attacks are mounted not by the primary user of a model but instead by a third party.” View source (opens in a new tab)
Provider (of AI)
Under the EU AI Act, whoever develops an AI system or model and places it on the market or puts it into service under its own name or trademark.
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(3): “a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge” View source (opens in a new tab)
R
RAG (retrieval-augmented generation)
A technique that combines a generative AI model with search over a knowledge base: it first retrieves the relevant information and then generates the answer from it, citing the sources. See service: Generative AI, RAG and knowledge bases →
Reference definition — Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” (2020): “We explore a general-purpose fine-tuning recipe for retrieval-augmented generation (RAG) — models which combine pre-trained parametric and non-parametric memory for language generation.” View source (opens in a new tab)
Risk (in AI)
The combination of the probability that an AI system will cause a problem and the severity of its consequences. Managing it is the basis of AI governance. See service: AI consulting and strategy →
Reference definition — NIST, AI Risk Management Framework 1.0 (NIST AI 100-1, 2023), Section 1.1: “In the context of the AI RMF, risk refers to the composite measure of an event’s probability of occurring and the magnitude or degree of the consequences of the corresponding event.” View source (opens in a new tab)
RLHF (reinforcement learning from human feedback)
A technique for adjusting a model using people’s ratings of its responses, so that it follows instructions better and behaves in a more helpful way. See service: Generative AI, RAG and knowledge bases →
Reference definition — Ouyang et al., “Training language models to follow instructions with human feedback” (2022): “We then collect a dataset of rankings of model outputs, which we use to further fine-tune this supervised model using reinforcement learning from human feedback.” View source (opens in a new tab)
S
Shadow AI
Employees’ use of AI tools without the approval or oversight of the IT department, which creates risks for data security and privacy. Read more on our blog (in Spanish) →
Systemic risk
A risk that, because of the scale and capabilities of the most powerful general-purpose AI models, can have broad effects on the market or on society.
Reference definition — Regulation (EU) 2024/1689 (AI Act), Art. 3(65): “a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain” View source (opens in a new tab)
T
Token
The smallest unit of text a language model processes; it can be a word, part of a word, or a symbol. The cost and usage limits of models are usually measured in tokens.
Training data
The set of examples used to train an AI model. Its quality, quantity, and representativeness largely determine the model’s performance and biases.
Transformer
A neural network architecture based on the attention mechanism that makes it possible to process sequences of text efficiently. It is the foundation of today’s language models. See service: Generative AI, RAG and knowledge bases →
Reference definition — Vaswani et al., “Attention Is All You Need” (2017): “We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely.” View source (opens in a new tab)
Trustworthy AI
AI that has the qualities needed to be used with confidence: it works well, is safe and transparent, respects privacy, and treats people fairly. See service: AI consulting and strategy →
Reference definition — NIST, AI Risk Management Framework 1.0 (NIST AI 100-1, 2023), Section 3: “Characteristics of trustworthy AI systems include: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed.” View source (opens in a new tab)
V
Vector database
A database that stores embeddings and lets you search content by similarity of meaning, not just by matching words. It is a common component of RAG solutions. See service: Generative AI, RAG and knowledge bases →
Sources and methodology
The explanation of each term is Wasivia’s own. The reference definitions come from primary sources and were checked against the original documents in October 2026: Regulation (EU) 2024/1689 on artificial intelligence (the official English text published in the Official Journal of the EU), the OECD, UNESCO, NIST, ISO/IEC standards, and the research papers that introduced each technique. Quotations reproduce the original English wording; when we could not access the original text, we give a paraphrase and label it as such.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act) — official English text (EUR-Lex) (opens in a new tab)
- OECD — Recommendation of the Council on Artificial Intelligence (OECD/LEGAL/0449) (opens in a new tab)
- UNESCO — Recommendation on the Ethics of Artificial Intelligence (2021) (opens in a new tab)
- NIST — Artificial Intelligence Risk Management Framework (AI RMF 1.0) (NIST AI 100-1) (opens in a new tab)
- NIST — Generative Artificial Intelligence Profile (NIST AI 600-1) (opens in a new tab)
- NIST — Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations (NIST AI 100-2e2025) (opens in a new tab)
- ISO/IEC 42001:2023 — Artificial intelligence management system (opens in a new tab)
- ISO/IEC 22989:2022 — Artificial intelligence concepts and terminology (opens in a new tab)
Missing a term? We update this glossary regularly. Last updated: October 12, 2026. This content is for informational purposes only and is not legal advice.
