AI Governance

Agentic AI describes AI systems designed to pursue goals through iterative reasoning, planning, tool use, and adaptation rather than producing only a single passive response.

An AI agent is a software system that perceives context, reasons about a goal, selects actions, and uses tools or other systems to complete tasks with a defined degree of autonomy.

AI guardrails are technical and organisational controls that constrain how an AI system receives inputs, generates outputs, uses data, and takes actions.

An AI impact assessment is a structured process for identifying, analysing, evaluating, and documenting the effects an AI system may have on people, organisations, society, and the environment throughout its lifecycle.

An AI management system is an organisation-wide set of policies, roles, processes, controls, and continual-improvement practices used to govern the responsible development, provision, and use of AI systems.

AI risk management is the coordinated process of identifying, analysing, evaluating, treating, monitoring, and communicating risks arising from the design, development, deployment, and use of AI systems.

Algorithmic bias is a systematic pattern in an algorithmic system that can produce unfair, inaccurate, or disproportionately harmful outcomes for particular people or groups.

Consent Mode is a Google tag-platform feature that adjusts tag and app-SDK behaviour according to a user’s consent choices.

A customer data platform (CDP) is packaged software that creates persistent, unified customer profiles and makes them available to other systems.

Data exfiltration is the unauthorised transfer, disclosure, or removal of data from a system, network, application, or controlled environment.

Data governance is the system of decision rights, accountability, policies, standards, and controls used to manage data as an organisational asset.

Data lineage is a traceable record of where data originated, how it moved and changed, and where it is used across systems and processes.

Data quality is the degree to which data is accurate, complete, consistent, timely, valid, unique, and fit for its intended use.

The EU AI Act is Regulation (EU) 2024/1689, a risk-based legal framework governing the development, placing on the market, deployment, and use of artificial intelligence in the European Union.

Explainable AI comprises methods and system properties that help relevant people understand the basis, behaviour, or limitations of an AI system and its outputs.

Generative Engine Optimization is the practice of improving how accurately and visibly an entity or its content is retrieved, represented, attributed, and cited in responses produced by generative search and answer systems.

Human-in-the-Loop is a system design in which human judgement, review, feedback, or authorisation is deliberately incorporated into an automated or AI-enabled process.

ISO 31000:2018 is an international guideline for integrating risk management principles, a framework and a process into organisational governance and decision-making.

ISO/IEC 22989:2022 is an international standard that establishes terminology and describes concepts used across artificial intelligence technologies and applications.

ISO/IEC 23894:2023 is an international standard providing guidance on managing risks faced by organisations that develop, provide, deploy, or use AI systems.

ISO/IEC 27001:2022 is the international requirements standard for establishing, implementing, maintaining and continually improving an information security management system.

ISO/IEC 38507:2022 provides guidance to governing bodies on the implications of an organisation’s use of artificial intelligence.

The first international standard specifying requirements for establishing, implementing, maintaining, and continually improving an AI management system within organisations.

ISO/IEC 42005:2025 is the published international standard providing guidance for organisations that conduct impact assessments of artificial intelligence systems.

A knowledge graph is a structured representation of entities, their attributes and the relationships between them, organised for machine querying and reuse.

A large language model (LLM) is a language model trained on very large datasets and parameterised to process or generate language and related sequences.

Lead scoring is a method for assigning values to prospect attributes and behaviours to support prioritisation or routing against an agreed outcome.

LLM tokenisation is the conversion of text or other input into model-specific token identifiers that a large language model can process.

A model card is a structured document that describes an AI or machine-learning model’s intended uses, performance, evaluation conditions, limitations, and relevant governance information.

A natural-language-processing task that identifies and classifies named entities such as people, organisations, locations and dates in text.

The field of computing concerned with enabling systems to analyse, generate and interact through human language.

The NIST AI Risk Management Framework is a voluntary, rights-preserving and use-case-agnostic framework for helping organisations manage risks to individuals, organisations, and society across the AI lifecycle.

A class of attack against large language model applications in which malicious input is crafted to override, redirect, or extract information from the system's intended instructions.

An AI architecture that combines large language model generation with real-time retrieval from external knowledge sources to produce grounded, verifiable responses.

Search Generative Experience was Google’s experimental Search Labs designation for generative-AI features in Search; it is now a legacy or transitional term associated with the development of AI Overviews and subsequent AI search experiences.

The computational classification or scoring of opinions, attitudes or emotional polarity expressed in text or other data.

User-generated content is material created and shared by customers, community members, or other users rather than produced solely by the organisation publishing or featuring it.

Web accessibility means designing and developing websites and web tools so people with disabilities can perceive, understand, navigate, interact with, and contribute to them.

An internationally recognised set of guidelines published by the W3C that define how to make web content more accessible to people with disabilities, organised around four principles: perceivable, operable, understandable, and robust.

Zero-party data is information that a customer intentionally and proactively shares with an organisation, such as preferences, intentions, context, or desired recognition.