AIGP Glossary

57 terms defined · Last reviewed: 2026-07-08 · By Victor Humenhuk (AIGP certified)

This glossary covers the terms that actually appear in the IAPP Artificial Intelligence Governance Professional (AIGP) body of knowledge: the regulated roles under the EU AI Act, the risk tiers, the assessments (DPIA, FRIA, conformity assessment), the privacy-enhancing technologies, the machine learning vocabulary, and the trust attributes that frameworks such as the NIST AI RMF and ISO/IEC 42001 are built on. Each entry opens with a short working definition and then gives the context that makes the term useful: which module it sits in, which law or framework uses it, and which terms it is usually confused with.

The definitions are not lifted from a standards body or a vendor blog. They come from the AIGP study notes written by Victor Humenhuk, and they are worded the way he needed them worded for the material to stick: plain English, no circular definitions, and an explicit note wherever the law, the frameworks and everyday technical usage pull a word in different directions. This is independent study material and is not affiliated with, endorsed by or approved by the IAPP, ISO, NIST or any regulator.

Use it in two ways. While you are working through a module, look up any term you cannot restate in your own words and follow the link back to the topic page it belongs to, so it lands in context rather than as an isolated flashcard. Later on, work down the list and try to define each term out loud before opening it, because the ones you stumble over are your revision list. The comparison pages below handle the pairs that are easiest to mix up under time pressure.

A

• Accountability - Identified people/orgs answerable for an AI system's functioning and impacts; anchor of governance.
• Adversarial attack - Deliberately crafted inputs designed to fool a model into errors or unsafe behaviour.
• Agentic AI - Multiple AI agents carrying out full end-to-end workflows with significant autonomy in complex environments.
• AI governance - An organisation's approach to using laws, policies, frameworks, practices and processes to keep AI responsible, ethical …
• AI Verify - Singapore's tool validating AI systems against 11 ethics principles.
• Algorithmic Impact Assessment (AIA) - Assessment covering data issues and documenting the stakeholder group's decisions and who accepts risk; Canada publishes…
• Anonymisation - Removing items that could identify individuals, such as names and addresses; complete anonymisation is difficult.
• Article 22 - GDPR provision: a general prohibition (with three exceptions) on decisions based solely on automated processing with leg…

B

• Bias - Systematic error producing unfair outcomes; computational, cognitive and societal flavours.
• Brittleness - An AI performing successfully in one instance yet failing in another.

C

• Conformity assessment - Process showing a system meets legal/standards requirements; EU AI Act gate for high-risk before market.

D

• Data drift - Input data's statistical properties change over time vs training data; performance degrades.
• Data lineage - Tracking the flow of data over time - origin, how it changed, destination - used to trace root causes.
• Data minimisation - Data must be adequate, relevant and limited to what is necessary; avoid 'nice to have' data.
• Data poisoning - Corrupting the training data so the model learns wrong or malicious behaviour.
• Data provenance - Tracking and logging the history and origin of data (sources, processes, actors, methods) to ensure integrity and identi…
• Deployer - Role that uses AI per regulations and provider instructions, maintains human oversight, keeps logs and runs impact asses…
• Differential privacy - PET that blurs information within datasets so data stays meaningful but individuals can't be identified.
• Distributor - Any entity other than the provider or importer that makes AI available on the market through the supply chain.
• DPIA - Data Protection Impact Assessment - evaluates risk of high-risk or significant processing; must meet or exceed GDPR requ…

E

• EU AI Act - The 2024 landmark, risk-based, extraterritorial regulation governing how AI is used, not the technology itself.
• Expert system - Older AI mimicking a human expert in one field via a knowledge base, inference engine and user interface.
• Explainability - The degree someone can consistently predict a model's result; essential for fairness, privacy, reliability, robustness, …

F

• Federated learning - PET where local systems train a central model on their own datasets, aggregated centrally without exposing individual da…
• Fine-tuning - Further training a pre-trained model on a smaller, task/domain-specific dataset.

G

• GDPR - The EU General Data Protection Regulation, in effect since 2018; technology-agnostic global baseline for data protection…

H

• Hallucinations - GenAI creating content that contradicts the source or is factually incorrect under the appearance of fact.
• Harms taxonomy - A list of negative consequences from data leak/misuse; an ontological map breaking harms into constituent components lik…
• Homomorphic encryption - A PET allowing computation on encrypted data - not yet at scale.

I

• Impact assessment - A risk management tool assessing an AI system's benefits, risks and limitations across the life cycle.
• Importer - Entity bringing an AI system into the domestic market from a third country; under the EU Act must be established in the …
• Interpretability - Degree a human can consistently understand and predict how the model works internally.
• ISO/IEC 42001 - AI management system standard underpinning foundational guardrails.

M

• Model card - Standardised record of a model's features, data, versions, bias/explainability reports, intended use and performance met…

N

• Neural networks - Brain-inspired algorithm for highly complex tasks like NLP and facial recognition.
• NIST AI RMF - US National Institute of Standards and Technology AI Risk Management Framework; nonbinding but widely referenced in proc…
• NYC Local Law 144 - Mandates AI bias audits for automated employment decision tools.

O

• OECD AI Principles - Five principles for trustworthy AI: inclusive growth; human rights & democratic values; transparency & explainability; r…
• Overfitting - Model learns the training examples too specifically and performs weakly on new, unseen data.
• Oversight - Supervision arrangements, human or institutional, that monitor, review and can intervene.

P

• PETs - Privacy-enhancing technologies; digital solutions letting data be used while protecting confidentiality, preventing inte…
• Privacy by design - Apply data protection from the initial planning stage and process by default only the data necessary for each purpose.
• Provider - Role ensuring safety/transparency before market, preparing technical documentation and carrying liability; the EU AI Act…

R

• RAG - Retrieval-augmented generation: pulls in external information to boost LLM accuracy and relevance.
• Recital 26 - GDPR recital emphasising pseudonymisation and anonymisation to safeguard personal data.
• Red teaming - Testing security by simulating adversarial attacks to expose flaws before public release.
• Reinforcement learning - An agent learns by trial and error in an environment, maximising reward via reinforcement and penalties.
• Reliability - Consistent, correct performance over time and across real-world conditions.
• Robustness - Performance maintained under stress: noisy, unexpected, adversarial or shifting inputs.

S

• Semi-supervised learning - A small labelled set steers learning over a large unlabelled set; course files LLMs here.
• Supervised learning - Learning from labelled data to predict outputs for new data; labelling is costly and can introduce bias.

T

• TEVV - Testing, Evaluation, Verification and Validation - continuous activity across the whole AI life cycle.
• Training data - The dataset the model learns from; must be representative, fair and compliant.
• Transparency - Making information available to stakeholders: a system's existence, capabilities, data, logic and limits.
• Trustworthy AI - AI that is lawful, ethical and robust: valid, safe, fair, accountable, transparent, privacy-protective.

U

• Underfitting - Model is too simple, misses important patterns, gives poor performance.
• Unsupervised learning - Learning from unlabelled data to find hidden patterns; cheaper but less accurate and subjective.

Keep going

The full AIGP study guide · Free practice questions · All study notes