AIGP key terms: the glossary and flashcards

The first performance indicator in the AIGP's Body of Knowledge is "Know the generally accepted definitions and types of AI", and the IAPP publishes the definitions it means for free: Key Terms for AI Governance, a glossary of about a hundred terms on its website and as a PDF. The certification FAQs point candidates to it directly. This guide sets out what the glossary is, the terms each of the exam's four domains leans on with the IAPP's own definitions, the places where a law defines the same thing differently, and how to learn a hundred definitions so that they hold up in a scenario question rather than only on a card.

The short answer

• What it is: the IAPP's own glossary for AI governance, free, first published in October 2023 and updated several times since. The web page counts 99 terms; the PDF of July 2025 indexes 101.
• Why it matters: the Body of Knowledge opens with definitions, and the FAQs say "For AI governance terms, review our resource Key Terms for AI Governance."
• What it is not: the law. In a question about the EU AI Act, the Act's definition decides, and it is worded differently.
• How to learn it: in small sets, by domain, with active recall, then in scenarios, because the handbook warns that the exams are not "a reflection of the candidate's ability to recall by memory the preparation materials".

What is the IAPP's AI governance glossary?

The glossary introduces itself: "This glossary was originally published in October 2023 and has been updated several times since to reflect the developing state of the art for AI governance. The updated version has been developed with reference to various materials and with valuable feedback from top experts in AI governance. It includes new terms and modifications to existing terms."

It is separate from the IAPP's privacy vocabulary: "Although there are some shared terms and definitions, this glossary is separate from the official IAPP Glossary of Privacy Terms." Three terms carry a footnote saying so: Accountability, Fairness and Transparency are each marked "Different definition than IAPP privacy training glossary." For a privacy professional, those three are the ones to relearn.

The web page shows its 99 terms 20 at a time; the PDF prints them in a two-column table with an index of 101, from Accountability to Weights. The handful of extra terms in the PDF do not change the core, and every definition quoted below appears in the PDF.

The terms behind Domain I

Domain I, the foundations, starts with the vocabulary. These are the definitions it leans on.

• AI governance: "A system of frameworks, practices and processes at an organizational level. AI governance helps various stakeholders implement, manage and oversee the use of AI technology."
• Artificial intelligence: "Artificial intelligence is a broad term used to describe an engineered system that uses various computational techniques to perform or automate tasks." The definition adds that "It may include automated decision-making."
• Machine learning: "Machine learning implements various algorithms that learn and improve by experience in a problem-solving process that includes data collection and preparation, feature engineering, training, testing and validation."
• Foundation model: "A large-scale model that has been trained on extensive and diverse datasets to enable broad capabilities". The glossary notes it is "Also called general purpose AI model and frontier AI".
• Agentic AI: "Agentic AI are AI systems that are designed to autonomously make and act upon decisions with limited human oversight." The Body of Knowledge names agentic AI in its introduction and "agentic architectures" in Domain IV.
• Hallucinations: "Instances in which generative AI models create seemingly plausible but factually incorrect outputs under the appearance of fact."
• Deepfakes: "Audio or visual content that has been altered or manipulated using generative artificial intelligence."
• Accountability: "The obligation and responsibility of the developers, deployers, and distributors of an AI system to ensure the system operates in a manner that is ethical, fair, transparent and compliant with applicable rules and regulations".
• Fairness: "An attribute of an AI system that prioritizes relatively equal treatment of individuals or groups in its decisions and actions in a consistent, accurate and measurable manner."
• Transparency: "The term may refer to the extent to which information regarding an AI system is made available to stakeholders, including disclosing if AI is used through techniques like watermarking, and explaining how the model works through model or system cards, for example."
• Explainability and interpretability: explainability is "The ability to describe or provide sufficient information about how an AI system generates a specific output or arrives at a decision in a specific context to a predetermined addressee." Interpretability is "The ability to explain or present a model's reasoning in human-understandable terms", and the glossary draws the line between them: "Unlike explainability, which provides an explanation after a decision is made, interpretability emphasizes designing models that inherently facilitate understanding through their structure, features or algorithms."
• Trustworthy AI: "In most cases, the term is used interchangeably with responsible AI and ethical AI, which all refer to principle-based AI development and governance".

The pair most often confused in practice questions is explainability and interpretability, and the glossary's own contrast, after the decision versus by design, is the cleanest way to keep them apart.

The terms behind Domain II

Domain II is the law, and the glossary supplies the vocabulary the laws assume.

• Automated decision-making: "The process of making a decision by technological means without human involvement."
• Conformity assessment: "An analysis, often performed by an entity independent of a model developer, on an AI system to determine whether requirements, such as establishing a risk management system, data governance, record-keeping, transparency and cybersecurity practices have been met."
• Contestability: "The principle of ensuring AI systems and their decision-making processes can be questioned or challenged by humans." The glossary adds that it is "Also called redress".
• Impact assessment: "An evaluation process designed to identify, understand and mitigate the potential ethical, legal, economic and societal implications of an AI system." Note the alias it gives: "Also referred to as a risk assessment."
• Oversight: "The process of effectively monitoring and supervising an AI system to minimize risks, ensure regulatory compliance and uphold responsible practices."

The terms behind Domain III

Domain III governs development, and its vocabulary is about data, testing and documentation.

• Data provenance: "A process that tracks and logs the history and origin of records in a dataset, encompassing the entire life cycle from its creation and collection to its transformation to its current state." Competency III.B asks candidates to "Establish and document data lineage and provenance."
• Data poisoning: "An adversarial attack in which a malicious user injects false data into a model to manipulate the training process, thereby corrupting the learning algorithm."
• Synthetic data: "Data generated by a system or model (see also machine learning model) that can mimic and resemble the structure and statistical properties of real data."
• Model card: "A brief document that discloses information about an AI model, like explanations about intended use, performance metrics and benchmarked evaluation in various conditions, such as across different cultures, demographics or race (see also system cards)." The Body of Knowledge lists "creating the model card" in competency III.C.
• System card: "Similar to a model card, a system card is also a brief transparency artifact, but instead of disclosing information about a model, it discloses information about how various AI models work together within a network of AI systems, promoting greater explainability of the system."
• Red teaming: "The process of testing the security of an AI system through an adversarial lens by removing defender bias."
• AI audit: "A systematic review of an AI system or a portfolio of AI systems to ensure it operates as intended and complies with relevant laws, regulations and standards."
• Watermarking: "The process of embedding subtle or visually imperceptible patterns in AI-generated content or metadata that can only be detected by computers."

The terms behind Domain IV

Domain IV governs deployment and use, and its terms are about adapting a model, keeping a human in control and staying stable after release.

• Fine-tuning: "The process of taking a foundation model and training it further for a specialized task through supervised learning."
• Retrieval augmented generation: "A technique applied to generative AI systems that allows the system to incorporate external information into a generated response." The Body of Knowledge names "retrieval augmented generation" among the deployment options in competency IV.A.
• Open-source software: "A development model that provides free and open access to source code to the public, which can then be viewed, modified and redistributed according to the terms of its respective license."
• Prompt engineering: "Prompt engineering is the deliberate process of structuring a prompt or series of prompts to influence model behavior to generate more desirable outputs and minimize harm."
• Human-in-the-loop: "A design paradigm that incorporates human oversight, intervention, interaction or control over the operation and decision-making processes of an AI system."
• Data drift: "The shift that occurs when a dataset being used as an input for an AI system changes over time so that it does not statistically resemble the dataset the model was trained on."
• Robustness and reliability: robustness is "An attribute of an AI system that signifies the system's ability to be resilient to, overcome and withstand security attacks"; reliability is "An attribute of an AI system that ensures it behaves as expected and performs its intended function consistently and accurately, even with new data that it has not been trained on."
• Fail-safe plans: "A back-up plan to be executed in case an AI system behaves in unexpected or dangerous ways."
• Shadow AI: "The unauthorized use of AI systems in an organization, often against organizational policies for data governance, privacy and AI use."

Where the glossary and the law differ

The glossary is the IAPP's vocabulary, not a statute, and Domain II questions turn on the statutes. Three differences are worth holding in mind.

What counts as AI. The glossary's artificial intelligence is "a broad term" for an engineered system using computational techniques. The EU AI Act's Article 3(1) is narrower and built for legal scope: "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". The OECD's definition shares that core: an AI system "is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs". In a question about whether the Act applies, inference is the word that decides, and it is not in the glossary's sentence.

Foundation models and general-purpose models. The glossary treats foundation model, general purpose AI model and frontier AI as one thing. The law does not: the AI Act regulates the "general-purpose AI model" with its own definition and duties, and California's SB 53 uses "frontier model" with a compute threshold. In a legal question, use the term the law uses.

Impact and risk assessments. The glossary says an impact assessment is "Also referred to as a risk assessment." Under the law, the names mark different instruments with different owners: the GDPR's data protection impact assessment, the AI Act's fundamental rights impact assessment for certain deployers, and the risk management system required of high-risk providers. A question that names one expects that one.

How to learn a hundred terms

The handbook sets the limit of memorizing: "Certification exams are not a reflection of the candidate's ability to recall by memory the preparation materials." The Body of Knowledge's verbs say the same. Domain I asks you to "Know" definitions, but many more indicators ask you to "Identify", "Understand", "Apply" or "Perform", so a definition has to survive a scenario.

Four habits help.

1. Learn by domain, not alphabetically. The glossary runs from A to W; the exam runs from foundations to deployment. The lists above are a starting order.
2. Recall, do not reread. Read the term, say the definition, then check. A flashcard is the simplest form of that.
3. Learn the pairs. Explainability and interpretability, robustness and reliability, model card and system card, data drift and data poisoning: exam options are often built from the neighbor of the right answer.
4. Move to scenarios early. Once a set is learned, answer questions that describe the situation without naming the term.

Our course follows that order. Its flashcards are written to the Body of Knowledge's competencies rather than copied from the glossary: 118 cards in the free first competency and 1,324 across the course, beside 1,364 practice questions that test the same terms in scenarios.

Where to start

The AIGP's first competency, AI and the need for governance, is free in full with a free account: five lessons, 100 questions and 118 flashcards covering the definitions and types of AI, its risks and harms, and the principles of responsible AI. AIGP free resources: what costs nothing lists the IAPP's other free documents, and AIGP exam questions: what they look like shows how definitions turn into questions.