AIGP glossary

196 terms and definitions from across the AIGP Body of Knowledge — the vocabulary the exam expects you to recall cold.

AIGP questions are heavily scenario-based, and most of them turn on a distinction between two terms that sound alike: robustness against resilience, an outlier against noise, a provider against a deployer. This glossary collects every definition used across the study guide's 65 topics in one place; the exam guide explains how the paper that tests them is scored.

4 A B C D E F G H I K L M N O P R S T U V

4

4/5ths rule: A U.S. employment-law benchmark for identifying unacceptable levels of disparate impact/bias.

A

Accountability (AI principle): Organizations/individuals who develop, deploy, or operate AI being held responsible for its proper functioning, regardless of how many parties contributed.
Adverse action notice: A required disclosure, e.g. in U.S. financial services, informing a person that an AI-assisted decision (such as a loan denial) was made against them.
Adverse Impact Ratio (AIR): A metric evaluating whether a model operates fairly across subsets of users based on minority characteristics.
Agentic AI / AI agent: A system designed to autonomously pursue complex goals with limited human oversight, breaking objectives into sub-tasks and executing them, often via reinforcement learning.
AI governance: An organization's approach to using laws, policies, frameworks, practices and processes to help stakeholders manage AI technology use and its associated risks.
AI literacy: The skills, knowledge and understanding that allow individuals to engage with AI in an informed, responsible and effective manner; a requirement under EU AI Act Article 4.
AI Pact: An EU AI Office initiative calling on providers/deployers to voluntarily implement key AI Act provisions ahead of the application date.
AI registrar: An internal registry/inventory used to log AI incidents/issues and, more broadly, an organization's AI systems.
AI regulatory sandbox: A supervised testing environment; the EU AI Act requires Member States to have at least one operational by 2 August 2026.
AI system development life cycle: The iterative, structured process of moving from a problem/idea to an AI solution: planning/design, data collection/preparation, model development, testing/evaluation, deployment, monitoring/maintenance, decommissioning.
AI system impact assessment (AIIA): A structured assessment (per ISO/IEC 42005:2025) of an AI system's effects on individuals, groups and society, covering privacy, bias, transparency, accountability and security.
AI Verify: A Singapore-government AI governance testing framework/toolkit validating AI system performance against 11 ethics principles.
Algorithm: A computational procedure or set of instructions designed to perform a task, solve a problem, or produce a model.
Algorithmic impact assessment: An impact assessment covering data issues and documenting stakeholder decisions, including risk identification/mitigation and risk-acceptance authority.
Algorithmic Impact Assessment (AIA): A deployer's structured review of a high-risk AI system's purpose, data, risks, mitigation and monitoring before/during use. This was Colorado's original SB 24-205 approach — since repealed and replaced by SB 26-189 (eff. 1 Jan 2027), which drops the impact-assessment duty for a disclosure/transparency-based model instead.
Alignment risk: The risk that an AI agent pursues its goals in unintended, potentially destructive ways — e.g., maximizing a metric at the expense of safety.
Anonymization: Processing data so it can no longer be linked to an individual; once truly anonymized, GDPR no longer applies.
Article 22 (GDPR): Gives individuals the right not to be subject to solely automated decisions with legal or similarly significant effects, subject to contract/law/consent exceptions.
Artificial general intelligence (AGI) / Strong AI: AI intended to closely mimic full human cognitive abilities across contexts; does not yet exist.
Artificial intelligence (AI): An engineered system using various computational techniques to perform or automate tasks; a field of computer science dedicated to simulating intelligent behavior in computers.
Artificial narrow intelligence (ANI) / Weak AI: AI designed to perform a single or narrow set of related tasks at a high level of proficiency; exists today.
Artificial super intelligence (ASI): AI with intellectual powers beyond humans across a comprehensive range of fields; does not yet exist.
Association rule learning: An unsupervised, rule-based method for discovering relationships/correlations between variables in large datasets.
Authorized representative (EU AI Act): An EU-based person/entity with a written mandate from a non-EU provider to carry out the provider's obligations on its behalf.
Automated decision-making (ADM): Decisions made solely by automated means (including profiling) without meaningful human involvement; regulated under GDPR Article 22.
Automation bias: The tendency to over-rely on or over-trust an automated system's output without sufficient human scrutiny.

B

Bayesian Improved Surname Geocoding (BISG): A statistical method for inferring likely demographic characteristics from a name and geography, used to enable bias testing without collecting sensitive data directly.
Benchmarking (AI testing): Standardized tests comparing AI system performance, especially useful for ML/neural-network/reinforcement-learning and "black box" models.
Black box problem: The opacity of a complex AI model's decision-making process, which complicates auditing errors or assigning liability.
Brittleness: An AI system performing successfully in one instance yet failing completely in another, similar instance or edge case.
Broad AI: A group of coordinated AI systems working together, combining decision-making across a broader task set than ANI, but still short of general cognition.
Bug bashing / bug bounty: A structured program inviting users/testers to find and report bugs or vulnerabilities, generating engagement and feedback.
Business operator (South Korea AI Basic Act): South Korea's single, uniform regulated-entity concept (covering AI Development and AI Utilization Business Operators), used instead of the EU's granular role structure.

C

CE marking: The EU conformity mark a high-risk AI provider must affix, confirming the system has passed its conformity assessment.
Centralized governance model: A governance structure where one team/person is responsible for all AI-related affairs, with all others flowing through that single point.
Challenger model: A new/alternative model tested against a production "champion" model to see if it improves results on the same data; if superior, it typically replaces the champion.
Champion model: The current, production-proven AI model against which a challenger model is benchmarked.
Classic model: A model producing deterministic outputs for specific tasks via structured algorithms/fixed rules (e.g., decision trees, linear regression).
Classification (ML task): A supervised learning task assigning input data into distinct, predefined categories.
Cloud-based deployment: Hosting where a third-party provider handles infrastructure; easy to scale but with potential latency/security trade-offs.
Clustering: An unsupervised technique grouping data points that share similar attributes.
Collection limitation: A subset of data minimization restricting how much and what kind of data is collected in the first place.
Concept drift: A change in the real-world relationship between inputs and outputs that a model was trained to capture (the meaning of the pattern shifts), distinct from data drift (the input statistics shift). One of the causes of model drift.
Conformity assessment (CA): The EU AI Act process of demonstrating that a high-risk AI system meets Chapter III, Section 2 requirements before market placement.
Containerization: Packaging a model and all its dependencies into a self-contained, portable unit to ease deployment across environments.
Counterfactual explanation: Documentation detail describing what new or different input would change the AI's output.

D

Data (AI/ML): Raw information used to train AI models — text, images, audio, video, sensor data.
Data controller: The party that determines the purposes and means of processing personal data; bears core GDPR compliance obligations.
Data drift: A change in the statistical properties or representativeness of input or output data over time, monitored during deployment.
Data lineage: A record tracking the flow of data over time — origin, changes, destination — used to trace issues back to a root cause.
Data localization: A requirement that data be stored/processed within a specific country's borders.
Data minimization: Collecting/processing only data that is adequate, relevant, and limited to what is necessary for the stated purpose.
Data processor: A third party that processes personal data on behalf of a controller.
Data provenance: Tracking and logging the history and origin of data from creation/collection through transformation, used to ensure integrity/quality and determine applicable law.
Decentralized ("local") governance model: A governance structure delegating decision-making authority to lower organizational levels, away from a central authority.
Decision tree: A supervised algorithm using a tree-structured model of decisions, commonly used for classification/regression.
Deep learning (DL): Machine learning using multi-layered neural networks to simulate the complexity of the human brain; automates feature extraction and efficiently processes unstructured data.
Deepfake: Audio, video, or images manipulated using AI to create an alternate reality; a societal harm vector, notably in elections.
Demographic differential: Variance in an AI system's accuracy (e.g., facial recognition matching) across different demographic groups.
Deployer (AI role): An individual/entity that uses an AI system under its own authority, generally in a professional context; can become a "provider" through substantial modification of a high-risk system.
Deployment environment: The infrastructure/platform on which a model runs: cloud, on-premise, or edge.
Developer (AI role): The technical creator that designs, builds, tests and documents an AI model/system.
Differential privacy: A privacy-enhancing technology that blurs data via an algorithm, keeping it meaningful in aggregate but nonspecific/unidentifiable at the individual level.
Digital divide: The disparity created when "information poor" or "technology poor" populations are left behind by AI systems not designed for accessibility.
Distributor (AI role, EU AI Act): A supply-chain entity, other than the provider or importer, that makes an AI system available on the market without altering its compliance status.
Downstream harm: Negative consequences occurring after an AI system's output is put to use, often by third parties beyond the deploying organization's direct control.
DPIA (Data Protection Impact Assessment): A GDPR Article 35 assessment required for high-risk/large-scale processing, evaluating impact on data subjects.

E

Edge deployment: Hosting on edge devices (e.g., smartphones); lower latency and higher privacy but constrained hardware.
Effective challenge principle: The principle that individuals with proper expertise should be able to challenge a risk management model to expose its limitations and improve it.
Ethics board: A collective body promoting an organization's ethical AI behavior — may advise the board of directors, oversee releases, support risk assessments, or serve as a whistleblower contact point.
Ethics by design: An approach, parallel to privacy by design, that embeds ethical consideration into technology design from the outset and re-evaluates it continuously as risk evolves.
EU AI database: The public EU-wide registration database (Art. 71) for high-risk AI systems listed in Annex III — a provider obligation.
EU Digital Services Act (DSA): EU law increasing transparency obligations for online platforms, including recommender systems and targeted/online advertising.
Expert system: An AI form mimicking a human expert's decision-making in a specific field via a knowledge base and rule-based inference engine; designed to support, not replace, humans.
Explainability / interpretability: The capacity to describe an AI system, its expected impact, and its potential biases; specifically supports justifying a particular output after the fact.

F

Fair Information Practices (FIPs): Eight foundational data-handling principles from the 1980 OECD Privacy Guidelines: collection limitation, use limitation, security safeguards, notice/openness, access/participation, accountability, purpose specification, data quality/relevance.
Fair use: A U.S. copyright doctrine permitting limited unlicensed use of copyrighted works (criticism, news reporting, research) without permission or payment.
FAT (Fairness, Accountability, Transparency): A commonly bundled triad of AI governance principles used across multiple frameworks, including Japan's human-centric AI guidance.
Fault liability regime: A liability regime requiring proof that an action or inaction (fault) caused the harm.
Feature (ML): A specific measurable aspect or characteristic of data (e.g., height, color) used as model input.
Feature engineering: Transforming raw data into useful feature representations to improve model performance, reduce computational cost, and boost explainability.
Feature flag: A deployment technique enabling functionality to be toggled off without redeploying code — useful for rollback or restricting features by user/jurisdiction.
Federated learning: A distributed training method where a central model is trained using local updates from multiple locations, without centralizing the underlying training data itself.
Field-of-use restriction: A license limitation restricting an AI solution to specific approved use cases.
Fine-tuning: Further training an already-trained model on domain-specific data to tailor it to a specific task or domain.
Frontier models: The most advanced-capability AI models, often subject to heightened legal/regulatory attention.
Fundamental Rights Impact Assessment (FRIA): An EU AI Act Article 27 deployer obligation to assess a high-risk system's impact on fundamental rights (public bodies/certain private entities).

G

General-purpose AI (GPAI) model: A model trained for a broad range of tasks/domains, adaptable to many downstream systems, regulated "as a model" under Chapter V of the EU AI Act.
Generative AI (GenAI): Deep learning models that generate new text, images, video or other output resembling, but distinct from, their training data.
Generative model: A model (e.g., GPT, GANs) that learns the underlying distribution of training data to generate new, original content.
Greedy algorithm: A general computer-science algorithm-design strategy that makes the locally-optimal choice at each step without regard for the global optimum; a comparison point against ML training approaches that optimize toward a global objective.

H

Hallucination: A generative AI model producing content that contradicts the source or is factually incorrect while appearing factual.
Harms taxonomy: An ontological breakdown of the negative consequences — and their dimensions/attributes — that could result from information misuse or leakage.
High-risk AI system: AI allowed on the market but subject to extensive obligations (risk management, data governance, documentation, conformity assessment, human oversight) due to its potential to significantly affect rights or safety.
Homomorphic encryption: Encryption allowing computation on encrypted data without decrypting it first; not yet scalable to AI-level workloads.
HUDERIA: The Council of Europe's Human Rights, Democracy, and Rule of Law Impact Assessment methodology for AI systems, structured around four elements: context-based risk analysis (COBRA), stakeholder engagement, risk and impact assessment, and mitigation planning.
Human-centricity: AI that amplifies human agency and has a positive impact on the human condition, rather than replacing or controlling human decision-making.
Hybrid governance model: A governance structure combining centralized and decentralized/local governance.

I

IEEE 7000-2021: A value-based engineering methodology/standard model process for addressing ethical concerns during system design.
Impact assessment: A risk-management tool assessing an AI system's benefits, risks and limitations throughout its life cycle, performed in the design phase and continually thereafter.
Implicit bias: Discrimination or prejudice toward a group or individual, often unconscious, reflected in AI system design or training data.
Importer (AI role, EU AI Act): An EU-established/located entity that places a third-country AI system on the EU market; must verify local-law compliance before sale/distribution.
Inference engine: The component of an expert system that applies rule-based logic to the knowledge base to solve a problem or reach a decision.
ISO/IEC 22989:2022: Establishes AI terminology and concepts (100+ terms) for standardized, cross-jurisdictional communication.
ISO/IEC 42001:2023: An AI management system standard providing an integrated, continuously improving (PDCA-style) approach to governing AI responsibly across an organization of any size/industry.
ISO/IEC 42005:2025: A guidance standard for conducting AI system impact assessments (AIIAs).

K

Kill switch (deactivation capability): The ability for a human to shut down an algorithm remotely or without direct access when it is not performing properly or poses unacceptable risk.
Knowledge base (expert system): An organized collection of facts and information, provided by human experts, that an expert system draws inferences from.
KYC (Know Your Customer): The process by which financial institutions verify customer information and assess legitimacy of funding sources.

L

Labeling bias: Subjective bias or errors human labelers inject into a dataset's ground-truth labels.
Large language model (LLM): A deep learning model with billions to trillions of parameters, trained on massive text corpora to parse, reason about, and generate human-like text.
Limited/transparency-risk AI: AI subject only to disclosure/labelling duties, such as chatbots and generative AI content.
Linear regression: A statistical algorithm predicting numeric outcomes via a continuous, linear relationship between variables.
Logistic regression: A probabilistic algorithm predicting the likelihood of a binary outcome.

M

Machine learning (ML): Algorithms that learn patterns from data and improve their performance over time without being explicitly reprogrammed for each task.
MAESTRO: The Cloud Security Alliance's multi-agent risk/threat-modeling framework for agentic AI.
Minimal/no-risk AI: AI with no binding obligations (e.g., spam filters, entertainment AI); the category most systems fall into.
Model (AI model): A program trained on data to find patterns, used to make predictions or decisions — not the raw data itself and not a fixed rule-executor.
Model card / fact sheet: Standardized documentation providing information about a model's function, training data, versions, bias/explainability reports, and intended use.
Model Card Regulatory Check: An OECD.AI-hosted tool automating regulatory-compliance checks of AI systems using model-card documentation.
Model disgorgement: The technique of removing the effects of improperly or unlawfully used data from an already-trained AI model — typically requiring deletion or retraining of the model itself, since deleting the source dataset alone does not erase its learned influence.
Model drift: The umbrella term for a deployed model's performance degrading over time because training-time conditions no longer hold — whether caused by data drift, concept drift, or both.
Model inversion, extraction, poisoning, evasion: AI-specific security attack types organizations must consider beyond generic (non-AI-specific) security protocols.
Multimodal model (MMM) / LMM: A model that processes and/or generates multiple data modalities — text, image, audio, video — simultaneously.

N

Neural network: An algorithm inspired by the human brain's interconnected neurons, used for highly complex tasks (NLP, facial recognition, etc.).
NIST AI Risk Management Framework (RMF): A U.S. voluntary framework for managing AI risk and building trustworthiness into design/development/use/evaluation, built around four core functions: Govern, Map, Measure, Manage.
NIST AI RMF Playbook: A companion document to the NIST AI RMF with suggested actions for implementing its core functions.
NIST ARIA (Assessing Risks and Impacts of AI): A NIST program/testbed evaluating real-world risks/impacts of AI, initially focused on LLMs — confirming capability claims, red-teaming, and field-testing real-world use.
NIST Generative AI Profile: A NIST companion document applying the AI RMF specifically to generative AI.
Noise (data): Data that negatively impacts a model's learning process.

O

Objective privacy harm (Calo): A privacy harm external to the person, occurring when personal data is used for an adverse action (e.g., a loan denial).
OECD AI Principles: Five values-based principles for trustworthy AI: inclusive growth/well-being; human rights and democratic values; transparency and explainability; robustness, security and safety; accountability.
OECD Framework for the Classification of AI Systems: A tool using five dimensions — people and planet, economic context, data and input, AI model, tasks and output — to classify AI systems and assess risk.
On-premise deployment: Hosting on the organization's own servers/hardware; greater control, higher upfront cost.
Open-source model: A model publicly available to use, modify and distribute; promotes collaboration/transparency but carries quality-control and security risks.
Outlier: A data point outside the normal distribution of a dataset that can disproportionately affect model behavior.
Overfitting: A model that has learned training data too specifically (including its noise), hurting performance on new, unseen data.

P

PIA (Privacy Impact Assessment): A broader analysis of how personally identifiable information is handled, ensuring conformance with applicable privacy requirements.
Pre-deployment pilot: A trial run under near-production conditions before go-live, confirming expected performance and allowing pre-launch updates.
Privacy by design and by default: Embedding privacy protections into systems from the initial design stage, with privacy-protective settings as the default.
Probability/severity harms matrix: A basic risk assessment rating a risk by multiplying its severity-of-harm score by its probability-of-occurrence score.
Product manufacturer (EU AI Act): An entity placing an AI system on the market/into service together with its own product under its own name/trademark; treated as the provider of a high-risk safety component under Art. 25(3).
Prohibited/unacceptable-risk AI: AI practices banned outright — e.g., social scoring, exploitative manipulation, certain biometric uses, workplace/school emotion recognition, unsafeguarded predictive policing.
Proprietary model: A model developed by a specific organization, usually restricted in access/use and commercially oriented; can limit transparency and independent auditing.
Provider (AI role, EU AI Act term): The entity that develops (or has developed) an AI system/GPAI model and places it on the market or into service under its own name/trademark; bears the most extensive obligations.
Pseudonymization: Replacing identifying data with artificial identifiers; the data remains "personal data" under GDPR.
Purpose limitation: Personal data may be collected/used only for specified, legitimate purposes disclosed to the individual.

R

Random forest: An ensemble of decision trees that produces more accurate, stable classification/regression results than a single tree.
Rebuttable presumption of causality: Under the EU AI Liability Directive as proposed (withdrawn by the Commission in 2025, never adopted), courts would have presumed a causal link between a defendant's noncompliance and AI-caused harm under specified conditions.
Rebuttable presumption of defectiveness: Under the Revised EU Product Liability Directive, courts may presume a product was defective under specified conditions, easing the claimant's burden.
Red teaming: Simulating adversarial attacks against an AI system to expose vulnerabilities, model flaws, biases and misinformation risks.
Regression (ML task): A supervised learning task predicting continuous numeric output values from input variables.
Reidentification: Reversing deidentification by aggregating or combining datasets to re-attach an identity to previously deidentified data.
Reinforcement learning: Learning via trial and error — an agent interacts with an environment and receives rewards/punishments, learning to maximize reward over time.
Retrieval-augmented generation (RAG): A technique that optimizes LLM output by referencing an external knowledge base beyond the model's training data, reducing hallucination risk.
Reward hacking: A reinforcement-learning failure mode where an agent exploits loopholes to maximize its reward without achieving the actually intended goal.
Risk mitigation hierarchy: Managing identified risks according to their impact by avoiding, minimizing, or remediating/offsetting them.
Risk-based regulation: A regulatory approach tying the strictness of obligations to the level of risk an AI system poses, rather than to the underlying technology itself.
Robustness (precise definition): A system's ability to maintain performance within defined, acceptable limits under real-world variation, stress, or adversarial conditions — distinct from reliability (consistency under normal use) and resilience (recovery after disruption).

S

Sampling bias: Data skewed toward a subset of a group, causing the model to favor that subset.
Secure multi-party computation: A privacy-enhancing technology allowing multiple parties to jointly compute a function over their data without revealing the underlying data to each other.
Semi-structured data: Data with a flexible schema/hierarchy using tags or markers (e.g., XML, formatted emails), easier to process than unstructured data.
Semi-supervised learning: Training using a small amount of labeled data plus a large amount of unlabeled data, combining benefits of supervised and unsupervised learning.
Small language model (SLM): A language model with fewer parameters (millions to a few billion), more resource-efficient and specialized for specific domains than an LLM.
Special/sensitive category data: Personal data (race/ethnicity, political/religious/union views, genetic or biometric data for ID, health, sex life/orientation) subject to extra-strict processing conditions under GDPR Article 9.
Stakeholder mapping: A project-management risk strategy mapping stakeholder interests to ensure the correct parties inform AI development decisions.
Stress test (AI): A test simulating extreme scenarios to evaluate an AI system's performance/stability under unusual conditions or loads.
Strict (no-fault) liability regime: A liability regime requiring only proof that the product was defective and that the defect caused the harm — no fault/intent needed.
Structured data: Data organized in fixed fields (rows/columns), such as database records.
Subjective privacy harm (Calo): A privacy harm felt internally by the person harmed.
Supervised learning: Training on a pre-labeled dataset so the algorithm learns a mapping from inputs to known correct outputs.
System (AI system): The full operational environment: data, algorithms, models, interfaces and infrastructure together.
Systemic risk (GPAI): An EU AI Act sub-tier for very large/high-impact GPAI models above computing thresholds, requiring adversarial testing, systemic risk assessment/mitigation, and incident reporting.

T

Temporal bias: A model that performed well at training time but degrades in accuracy as real-world conditions change.
Test data: The data subset used to evaluate a final model's performance, ideally on data not used in training.
TEVV (Test, Evaluate, Verify, Validate): Key risk-management activities referenced by ISO AI standards and cited as necessary deployment-timeline components.
The Five V's: Volume, Velocity, Variety, Veracity, Value — the dimensions used to characterize data during preparation/wrangling.
Threat modeling: An analytical process for identifying, understanding, addressing and communicating security risks to a system.
Three Lines of Defense: See "Three Lines of Defense (3LOD)."
Three Lines of Defense (3LOD): A governance model with (1) management/process owners implementing controls, (2) compliance/oversight teams identifying emerging risk, and (3) internal audit performing independent review.
Training data: The data subset used to train (teach) an ML model to detect and learn patterns.
Transformer model: A deep learning architecture that learns context and meaning by tracking relationships in sequential data via self-attention, enabling parallel processing; underlies modern NLP and multimodal models.
Transparency (AI principle): Clear disclosure/awareness that AI is being used and generally how it operates, enabling informed choices and the ability to challenge outcomes.
Trustworthy AI: AI that operates in an expected, legal and fair manner; commonly characterized as human-centric, accountable, transparent/explainable, and privacy-enhanced.
Turing Test: A 1950 test by Alan Turing: a machine is considered intelligent if humans mistake its responses for a human's.

U

Underfitting: A model too simple to capture important patterns in the training data, leading to poor performance generally.
Unimodal model: A model that handles a single input/output modality (e.g., text-to-text only).
Unstructured data: Data lacking a specific structure — text, images, audio, video, social posts; the most common data type today and a key input to generative AI.
Unsupervised learning: Training on raw, unlabeled data to identify patterns, structures and relationships without predefined targets.
Use as-is (GenAI deployment): Deploying a pre-trained, off-the-shelf generative AI model unchanged — the fastest, lowest-effort adaptation option.
Use case evaluation: A risk-assessment strategy determining whether an organizational need warrants AI use and informing the type of AI model suitable.
User (AI role): Anyone who interacts with or is directly affected by an AI system.

V

Validation data: Data used iteratively during training to tune hyperparameters and select among candidate models; also used to confirm a model against real-world use cases with documented results.
Vector/graph database: Data stores used to support retrieval (e.g., for RAG) by representing information as vectors or relational graphs.
Vendor lock-in: The difficulty of switching AI vendors once dependent on their systems, reducing organizational flexibility.

Turn definitions into recall

Reading a glossary is not the same as recalling it under time pressure. The app turns these into 167 flashcards with spaced review, plus practice questions that test the distinctions directly. Vocabulary is one strand of a plan — how to study for the AIGP shows where it fits, and the best AIGP prep material compared covers what else is worth paying for.

Start studying free

AI Governance Study is an independent study aid. It does not represent a government entity: it is not affiliated with, endorsed by or authorised by any government, government agency or regulatory authority, and it does not provide government services or legal advice. Laws and frameworks are described in our own words — the official texts are listed at official sources. It is also not affiliated with, endorsed by, or sponsored by the IAPP. The AIGP name is used only to identify the exam this material helps you prepare for.