Clear, practical definitions of the technologies, concepts, risks, governance requirements and terminology shaping artificial intelligence.
Artificial intelligence introduces new technologies, terminology and governance concepts at extraordinary speed.
The INTAI AI Glossary explains important terms in clear business language and connects them to deeper guidance, assessments, frameworks and practical resources across the INTAI knowledge base.
Use it as a reference point for understanding AI, governing it responsibly and exploring the topics that matter to your organisation.
Search definitions directly or browse alphabetically to explore the technologies, risks, governance concepts and terminology shaping artificial intelligence.
Explore: AI Governance Policies · AI Governance Strategy
Acceptable AI use defines how employees and other authorised users may use artificial intelligence within an organisation, including which tools and activities are permitted, restricted or prohibited and what responsibilities apply when AI is used for work.
Explore: AI Governance Strategy · Governance Operating Models
Accountability means having clearly identified people or functions responsible for decisions, actions and outcomes, with the authority and obligation to explain how responsibilities have been fulfilled.
Explore: AI Agents Readiness Assessment · AI Lifecycle
Agentic AI refers to AI systems designed to pursue goals, plan activities, make decisions about next steps and take actions through connected tools, information or systems with varying levels of human involvement.
Explore: AI Agents Readiness Assessment
An agentic workflow is a process in which one or more AI agents can plan tasks, use tools, make intermediate decisions and take actions across multiple steps within defined permissions, boundaries and controls.
Explore: AI Governance Strategy · Governance Operating Models
AI accountability is the principle that clearly identified people or organisational roles remain responsible for significant decisions, actions and outcomes involving artificial intelligence, even where AI systems support or automate parts of the process.
Explore: AI Adoption Assessment
AI adoption is the process through which artificial intelligence moves from experimentation into sustained and useful application across an organisation's people, processes, workflows and operating practices.
Explore: AI Agents Readiness Assessment
An AI agent is an AI-enabled system that can pursue a defined goal, determine what actions to take and interact with tools, information or other systems with varying levels of autonomy.
Explore: Responsible AI
AI alignment concerns whether an AI system behaves in ways that remain consistent with its intended objectives, instructions, constraints and relevant human or organisational values.
Explore: AI Assurance Reviews · AI Assurance Assessment
AI assurance is the structured process of gathering and evaluating evidence to determine whether an AI system is operating as intended and whether its risks, controls and outcomes can be relied upon with an appropriate level of confidence.
Explore: AI Assurance Reviews
An AI audit is a structured examination of an AI system, process or governance environment to assess whether defined requirements are being followed and whether appropriate controls, evidence and oversight are in place.
Explore: Bias & Fairness Audits · Responsible AI
AI bias refers to systematic patterns in an AI system that can produce distorted, unfair or disproportionately harmful outcomes for particular individuals, populations or groups.
Explore: Responsible AI
AI ethics considers the moral principles, human values and wider impacts that should inform how artificial intelligence is designed, developed, deployed and used.
Explore: AI Governance Policies
AI-generated content is text, imagery, audio, video, software code or other material created wholly or substantially through the use of a generative AI system.
Explore: AI Governance Strategy · AI Governance Assessment
AI governance is the system of responsibilities, decision rights, policies, controls and oversight used to direct how artificial intelligence is selected, developed, procured, deployed, used, monitored and retired across an organisation.
Explore: AI Governance Policies · AI Governance Strategy
Acceptable AI use defines how employees and other authorised users may use artificial intelligence within an organisation, including which tools and activities are permitted, restricted or prohibited and what responsibilities apply when AI is used for work.
Explore: AI Governance Strategy · Governance Operating Models
Accountability means having clearly identified people or functions responsible for decisions, actions and outcomes, including responsibility for providing appropriate oversight, managing risks and addressing failures when they occur.
Explore: AI Agents Readiness Assessment · AI Lifecycle
Agentic AI refers to AI systems designed to pursue goals, plan activities, determine next steps and take actions through connected tools or systems with varying levels of human involvement and autonomy.
Explore: AI Agents Readiness Assessment
An agentic workflow is a process in which one or more AI agents can plan tasks, use tools, make intermediate decisions and take actions across multiple steps within defined permissions, controls and oversight arrangements.
Explore: AI Governance Strategy · Governance Operating Models
AI accountability is the principle that clearly identified people or organisational roles remain responsible for significant decisions, actions and outcomes involving artificial intelligence, even where AI systems or external providers contribute to those outcomes.
Explore: AI Adoption Assessment
AI adoption is the process through which artificial intelligence moves from experimentation into sustained and useful application across an organisation's people, processes, workflows and operating practices.
Explore: AI Agents Readiness Assessment
An AI agent is an AI-enabled system that can pursue a defined goal, determine what actions to take and interact with tools, information or other systems with varying levels of autonomy and human oversight.
Explore: Responsible AI
AI alignment concerns whether an AI system behaves in ways that remain consistent with its intended objectives, instructions, constraints and relevant human or organisational values.
Explore: AI Assurance Reviews · AI Assurance Assessment
AI assurance is the structured process of gathering and evaluating evidence to determine whether an AI system is operating as intended and whether its risks, controls and outcomes can be relied upon with an appropriate level of confidence.
Explore: Responsible AI
AI-assisted decision-making occurs when artificial intelligence provides information, analysis, predictions or recommendations that influence a decision while a person retains meaningful authority and responsibility for the final decision.
Explore: AI Assurance Reviews
An AI audit is a structured examination of an AI system, process or governance environment to assess whether defined requirements are being followed and whether appropriate controls, evidence and oversight are in place.
Explore: Bias & Fairness Audits · Responsible AI
AI bias refers to systematic patterns in an AI system that can produce distorted, unfair or disproportionately harmful outcomes for particular individuals, populations or groups.
Explore: Responsible AI
AI ethics considers the moral principles, values and wider human impacts that should inform how artificial intelligence is designed, developed, deployed and used.
Explore: AI Governance Policies
AI-generated content is text, imagery, audio, video, software code or other material created wholly or substantially through the use of a generative AI system.
Explore: AI Governance Strategy · AI Governance Assessment
AI governance is the system of responsibilities, decision rights, policies, controls and oversight used to direct how artificial intelligence is selected, developed, procured, deployed, used, monitored and retired across an organisation.
Explore: INTAI Governance Framework · INTAI Templates
An AI governance framework is the organised structure through which an organisation defines the principles, responsibilities, decision processes, controls and lifecycle requirements used to govern artificial intelligence.
Explore: AI Risk Assessment · INTAI Templates
An AI impact assessment is a structured evaluation of how an AI system could affect people, rights, decisions, safety, operations and other relevant organisational or societal interests.
Explore: AI Incident Management · AI Governance Policies
An AI incident is an event in which an AI system causes, contributes to or creates a significant risk of harmful, incorrect, unsafe, unauthorised or otherwise unacceptable outcomes.
Explore: INTAI Templates · AI Governance Policies
An AI inventory is a maintained record of the AI systems, tools and significant AI use cases operating across an organisation, typically including their purpose, ownership, risk classification, suppliers and lifecycle status.
Explore: AI Lifecycle Framework
The AI lifecycle describes the stages through which an AI system or use case moves, from initial idea and assessment through development or procurement, testing, deployment, operation, monitoring, change and eventual retirement.
Explore: AI Literacy Assessment · AI Training
AI literacy is the ability to understand what AI can and cannot do, recognise where it is being used, work with it appropriately and understand the risks, limitations and responsibilities associated with AI-assisted activity.
Explore: ISO/IEC 42001
An AI management system is the organised set of policies, objectives, responsibilities and processes through which an organisation manages artificial intelligence and continually improves its governance arrangements.
Explore: Executive AI Maturity Assessment
AI maturity describes how developed an organisation's capabilities are for leading, governing, adopting and realising value from artificial intelligence in a consistent and sustainable way.
Explore: Model Governance
An AI model is a computational system trained or configured to identify patterns and produce outputs such as predictions, classifications, recommendations or generated content from the information it receives.
Explore: AI Monitoring Services · AI Lifecycle
AI monitoring is the ongoing observation and review of an AI system's performance, behaviour, risks and control effectiveness after deployment so that deterioration, unexpected outcomes and emerging issues can be identified.
Explore: AI Opportunity Assessment
An AI opportunity is a business problem, process or unmet need where applying artificial intelligence could create measurable value, improve performance or enable a new capability.
Explore: AI Technology Implementation
AI orchestration is the coordination of models, agents, tools, information sources and workflows so they can operate together as part of a broader AI-enabled process or system.
Explore: AI Governance Policies · AI Policy Assessment
An AI policy is a formal organisational document that establishes rules, responsibilities, requirements and expectations for how artificial intelligence should be used, managed or controlled.
Explore: AI Readiness Assessment
AI readiness describes how prepared an organisation is to adopt and use artificial intelligence effectively, considering factors such as leadership, people, skills, data, technology, processes, opportunities and organisational change.
Explore: AI Assurance Reviews
AI red teaming is the deliberate testing of an AI system by attempting to expose unsafe behaviour, vulnerabilities, control weaknesses and potential misuse before they create real-world harm.
Explore: EU AI Act · AI Compliance Readiness
AI regulatory compliance is the process of identifying and meeting the laws, regulations and regulatory obligations that apply to an organisation's AI systems, activities and responsibilities.
Explore: AI Risk Assessment · AI Risk Assessment Tool
AI risk is the possibility that the development, deployment or use of artificial intelligence could create harmful, incorrect, unsafe, unlawful, unfair or otherwise undesirable outcomes for an organisation or its stakeholders.
Explore: AI Risk Classification · EU AI Act
AI risk classification is the process of assigning an AI system or use case to a defined risk level based on factors such as its purpose, potential impact, affected people, operational significance and applicable regulatory requirements.
Explore: Responsible AI · AI Assurance Reviews
AI safety is the discipline of preventing, detecting and reducing harmful, unintended or unsafe outcomes arising from the development and use of artificial intelligence.
Explore: AI Skills Assessment
AI skills are the practical capabilities people need to use, manage, evaluate, develop or oversee artificial intelligence effectively within their role.
Explore: AI Lifecycle · INTAI Templates
An AI system is a machine-based system that uses artificial intelligence techniques to generate outputs such as predictions, recommendations, decisions, classifications or content from the inputs it receives.
Explore: AI Training · AI Training Needs Assessment
AI training is structured learning designed to build the knowledge and practical capabilities people need to use, manage or oversee artificial intelligence appropriately.
Explore: Responsible AI · AI Governance Policies
AI transparency is the practice of providing appropriate information about where and how AI is being used, what role it plays and what limitations, risks or responsibilities affected people should understand.
Explore: AI Opportunity Assessment
An AI use case is a defined application of artificial intelligence to a particular business problem, opportunity, task or process with an identified purpose and intended outcome.
Explore: AI Opportunity Assessment
AI value realisation is the process of converting AI investment and adoption into measurable business outcomes such as increased revenue, lower costs, improved productivity, better decisions or enhanced service.
Explore: AI Vendor Assessments · AI Vendor Assessment Tool
An AI vendor is an external organisation that provides an AI model, platform, application, service or AI-enabled technology used by another organisation.
Explore: AI Training · AI Skills Assessment
AI workforce impact describes how artificial intelligence changes jobs, tasks, skills, responsibilities, working practices and workforce requirements across an organisation.
An algorithm is a defined set of rules or computational instructions used to process information and produce a result. Algorithms are fundamental to software and are used extensively within artificial intelligence and machine learning systems.
Explore: Responsible AI
Algorithmic accountability is the principle that organisations remain responsible for the design, use and consequences of algorithmic systems and should be able to demonstrate appropriate oversight, ownership and controls.
Explore: Bias & Fairness Audits · Responsible AI
Algorithmic bias refers to systematic patterns within an algorithm or its application that can produce distorted, unfair or disproportionately harmful outcomes for particular individuals or groups.
Explore: Explainability Reviews
Algorithmic transparency is the provision of appropriate information about how an algorithmic system is used, what role it plays and how its operation or outcomes can be understood and challenged.
Explore: Responsible AI
Alignment describes the extent to which an AI system's behaviour and outputs remain consistent with intended objectives, organisational requirements, human values and defined constraints.
Anomaly detection is the use of analytical or machine learning techniques to identify data, events or behaviour that differs significantly from expected patterns and may indicate an error, threat, failure or other issue requiring attention.
Explore: AI Technology Implementation
An application programming interface, or API, is a defined way for software systems to communicate and exchange information. APIs are commonly used to connect AI models and services with applications, data sources and business systems.
Artificial general intelligence, or AGI, is a proposed form of AI capable of performing a broad range of intellectual tasks at a level comparable to or beyond human capability. There is no universally agreed definition or test for when AGI has been achieved.
Explore: INTAI · AI Assessments
Artificial intelligence, or AI, refers broadly to computer systems designed to perform tasks associated with aspects of human intelligence, including understanding language, recognising patterns, making predictions, generating content, solving problems and supporting decisions.
An artificial neural network is a machine learning architecture made up of interconnected computational units arranged in layers. Neural networks can learn complex patterns from data and underpin many modern AI systems.
Explore: Responsible AI · AI Governance Policies
Automated decision-making occurs when a system makes or materially determines a decision with limited or no direct human intervention.
Automation is the use of technology to perform tasks or processes with reduced human intervention. Automation does not necessarily involve AI, although AI can enable more adaptive and complex forms of automated activity.
Explore: Responsible AI
Automation bias is the tendency for people to place excessive trust in automated recommendations or outputs and fail to apply sufficient independent judgement or challenge.
Explore: AI Agents Readiness Assessment
Autonomous AI refers to AI systems capable of making decisions or taking actions with limited direct human intervention. The degree of autonomy can vary significantly and should influence the level of governance and oversight applied.
A benchmark is a defined test, dataset or reference point used to compare the performance, behaviour or capabilities of models and systems.
Explore: Bias & Fairness Audits
Bias is a systematic tendency that can influence data, models, processes or decisions and lead to distorted, inaccurate or unfair outcomes.
Explore: Bias & Fairness Audits
Bias testing is the structured evaluation of an AI system or model to identify whether its performance or outcomes differ unfairly across relevant individuals, populations or groups.
Explore: Explainability Reviews
A black box AI system is one whose internal processing or decision process is difficult for people to understand or explain, even when its inputs and outputs can be observed.
Explore: Board Advisory · Board AI Readiness Assessment
Board AI oversight is the governance exercised by a board over an organisation's material AI strategy, investment, risk, regulatory exposure and accountability, while preserving the distinction between board oversight and executive management.
Explore: AI Opportunity Assessment
An AI business case sets out why a proposed AI investment or use case should proceed, including the problem being addressed, expected value, costs, risks, dependencies, implementation requirements and measures of success.
Chain of thought commonly refers to intermediate reasoning steps associated with an AI system working through a problem, although a displayed explanation should not automatically be treated as a complete record of a model's internal processing.
A chatbot is a software application designed to interact with people through written or spoken conversation. Modern chatbots may use generative AI and large language models to provide more flexible and context-aware responses.
Explore: ChatGPT Readiness Assessment · AI Governance Policies
ChatGPT is an AI service developed by OpenAI that enables people to work with AI models through natural-language interaction and tools supporting activities such as writing, analysis, research, coding and problem-solving.
A classification model is a machine learning model designed to assign data or observations to defined categories, such as identifying whether a transaction is potentially fraudulent or classifying a message by topic.
Explore: AI Governance Policies
Claude is a family of generative AI models and an AI assistant developed by Anthropic, designed to support natural-language tasks including writing, analysis, reasoning and coding.
Explore: AI Technology Implementation · AI Vendor Assessments
Cloud AI refers to artificial intelligence models, platforms and services delivered through cloud computing infrastructure rather than operating entirely within an organisation's own local technology environment.
Explore: AI Compliance Readiness
Compliance is the process of meeting applicable laws, regulations, contractual requirements, standards and internal policies.
Explore: AI Technology Implementation
Computer vision is a field of artificial intelligence that enables systems to analyse and interpret visual information such as images and video, including recognising objects, detecting patterns and identifying events.
Explore: AI Monitoring Services
Concept drift occurs when the relationship between the information a model receives and the outcome it is intended to predict changes over time, potentially reducing the model's continuing reliability.
A confidence score is a numerical indication of how strongly a model supports a particular prediction or classification. It should not automatically be interpreted as a guarantee that the output is correct.
Explore: EU AI Act · AI Compliance Readiness
A conformity assessment is a process used to determine whether a product, system or service meets specified legal or technical requirements. Under the EU AI Act, certain high-risk AI systems are subject to conformity assessment requirements before being placed on the market or put into service.
Content moderation is the process of identifying, reviewing, restricting or removing content that breaches defined rules or standards. AI can support moderation, although significant decisions may require human review and clear escalation processes.
A context window is the amount of information an AI model can consider within a single interaction or processing sequence. In language models, this commonly includes the user's input, instructions, conversation history and other supplied content.
Explore: INTAI Governance Framework
A control is a measure designed to prevent, detect, reduce or respond to risk. AI controls can include approvals, access restrictions, testing, human review, monitoring, documentation and escalation requirements.
Explore: AI Assurance Reviews
Control effectiveness describes whether a control is appropriately designed, implemented and operating in practice to manage the risk it is intended to address.
A co-pilot is a general term for an AI assistant designed to work alongside a person by supporting tasks, generating suggestions or helping complete activities while leaving meaningful responsibility with the human user.
Copilot is a name used by Microsoft for a range of AI-assisted products and capabilities. Microsoft 365 Copilot integrates generative AI with Microsoft 365 applications and organisational information, subject to the user's permissions and the configuration of the organisation's environment.
Explore: AI Governance Policies
Copyright and AI refers to the legal and governance questions that arise when copyrighted material is used in connection with AI systems or when AI is used to generate, transform or reproduce content. Applicable rights and obligations depend on the circumstances and relevant jurisdiction.
Data anonymisation is the process of altering information so that individuals can no longer be identified by reasonably available means. Effective anonymisation is different from pseudonymisation, where re-identification remains possible using additional information.
Explore: AI Monitoring Services
Data drift occurs when the characteristics or distribution of data encountered by an AI system change over time compared with the data used to develop, train or validate it, potentially affecting performance.
Explore: AI Data Readiness Assessment
Data governance is the system of responsibilities, standards and controls used to manage data throughout its lifecycle, including its ownership, quality, access, security, use, retention and disposal.
Explore: AI Data Readiness Assessment
Data lineage records how data moves through systems and processes, including where it originated, how it was transformed and where it has been used.
Explore: AI Governance Policies
Data minimisation is the principle that organisations should collect, use and retain only the personal data that is adequate, relevant and necessary for the defined purpose.
Data poisoning is the deliberate manipulation or contamination of data used by an AI or machine learning system with the aim of altering its behaviour, reducing its performance or creating hidden vulnerabilities.
Explore: AI Governance Policies
Data privacy concerns how personal information is collected, used, shared, retained and protected, including the rights and expectations of individuals whose information may be processed by AI systems.
A data protection impact assessment, or DPIA, is a structured process for identifying and reducing privacy risks associated with processing personal data, particularly where processing is likely to create high risks to individuals.
Explore: AI Data Readiness Assessment
Data provenance is information about where data originated, how it was collected, how it has changed and how it has been used, helping organisations assess its suitability, reliability and governance.
Explore: AI Data Readiness Assessment
Data quality describes whether data is sufficiently accurate, complete, consistent, relevant, timely and reliable for its intended purpose. Poor-quality data can materially affect AI performance and outcomes.
Explore: AI Data Readiness Assessment
Data readiness describes whether an organisation has data of sufficient quality, accessibility, governance and suitability to support a proposed AI initiative or capability.
A dataset is an organised collection of information used for analysis or computational processing. In machine learning, datasets may be used to train, validate, test or monitor models.
Explore: Governance Operating Models
Decision rights define which individuals, roles or governance bodies have authority to make, approve, challenge or escalate particular decisions.
Deep learning is a branch of machine learning that uses neural networks with multiple computational layers to learn complex patterns from large amounts of data.
Explore: AI Training · Responsible AI
A deepfake is synthetic or manipulated media created using AI to convincingly imitate or alter a person's appearance, voice or actions, potentially making fabricated content appear authentic.
Explore: EU AI Act
Under the EU AI Act, a deployer is generally a person or organisation using an AI system under its authority in a professional context, subject to the specific definitions and exclusions set out in the legislation.
Explore: AI Lifecycle
Deployment is the stage at which an AI system is introduced into a live environment where it can be used within real organisational processes, products or services.
Explore: AI Monitoring Services
Drift is a change over time in the data, environment or relationships on which an AI model depends, potentially causing its behaviour or performance to deteriorate or become less appropriate.
An artificial neural network is a machine learning architecture made up of interconnected computational units arranged in layers. Neural networks can learn complex patterns from data and underpin many modern AI systems.
Explore: Responsible AI · AI Governance Policies
Automated decision-making occurs when a system makes or materially determines a decision with limited or no direct human intervention.
Automation is the use of technology to perform tasks or processes with reduced human intervention. Automation does not necessarily involve AI, although artificial intelligence can enable more adaptive and complex forms of automated activity.
Explore: Responsible AI
Automation bias is the tendency for people to place excessive trust in automated recommendations or outputs and fail to apply sufficient independent judgement, verification or challenge.
Explore: AI Agents Readiness Assessment
Autonomous AI refers to AI systems capable of making decisions or taking actions with limited direct human intervention. The degree of autonomy can vary significantly and should influence the level of governance, permissions and oversight required.
A benchmark is a defined test, dataset or reference point used to compare the performance, behaviour or capabilities of models and systems under specified conditions.
Explore: Bias & Fairness Audits
Bias is a systematic tendency that can influence data, models, processes or decisions and lead to distorted, inaccurate or unfair outcomes.
Explore: Bias & Fairness Audits
Bias testing is the structured evaluation of an AI system or model to identify whether its performance or outcomes differ unfairly across relevant individuals, populations or groups.
Explore: Explainability Reviews
A black box AI system is one whose internal processing or decision process is difficult for people to understand or explain, even when its inputs and outputs can be observed.
Explore: Board Advisory · Board AI Readiness Assessment
Board AI oversight is the governance exercised by a board over an organisation's material AI strategy, investment, risk, regulatory exposure and accountability, while preserving the distinction between board oversight and executive management.
Explore: AI Opportunity Assessment · INTAI Templates
An AI business case sets out why a proposed AI investment or use case should proceed, including the problem being addressed, expected value, costs, risks, dependencies, implementation requirements and measures of success.
Chain of thought is a term commonly used to describe intermediate reasoning steps associated with an AI system working through a problem. A displayed explanation should not automatically be treated as a complete or literal record of a model's internal processing.
A chatbot is a software application designed to interact with people through written or spoken conversation. Modern chatbots may use generative AI and large language models to provide more flexible and context-aware responses.
Explore: ChatGPT Readiness Assessment · AI Governance Policies
ChatGPT is an AI service developed by OpenAI that enables people to work with AI models through natural-language interaction and tools supporting activities such as writing, analysis, research, coding and problem-solving.
A classification model is a machine learning model designed to assign data or observations to defined categories, such as identifying whether a transaction is potentially fraudulent or classifying a message by topic.
Explore: AI Governance Policies
Claude is a family of generative AI models and an AI assistant developed by Anthropic, designed to support natural-language tasks including writing, analysis, reasoning and coding.
Explore: AI Technology Implementation · AI Vendor Assessments
Cloud AI refers to artificial intelligence models, platforms and services delivered through cloud computing infrastructure rather than operating entirely within an organisation's own local technology environment.
A co-pilot is a general term for an AI assistant designed to work alongside a person by supporting tasks, generating suggestions or helping complete activities while leaving meaningful responsibility with the human user.
Explore: AI Compliance Readiness
Compliance is the process of meeting applicable laws, regulations, contractual requirements, standards and internal policies.
Explore: AI Technology Implementation
Computer vision is a field of artificial intelligence that enables systems to analyse and interpret visual information such as images and video, including recognising objects, detecting patterns and identifying events.
Explore: AI Monitoring Services
Concept drift occurs when the relationship between the information a model receives and the outcome it is intended to predict changes over time, potentially reducing the model's continued effectiveness.
A confidence score is a numerical indication of how strongly a model supports a particular prediction or classification. It should not automatically be interpreted as a guarantee that the output is correct.
Explore: EU AI Act · AI Compliance Readiness
A conformity assessment is a process used to determine whether a product, system or service meets specified legal or technical requirements. Under the EU AI Act, certain high-risk AI systems are subject to conformity assessment requirements.
Content moderation is the process of identifying, reviewing, restricting or removing content that breaches defined rules or standards. AI can support moderation, although significant decisions may require human review and clear escalation processes.
A context window is the amount of information an AI model can consider within a single interaction or processing sequence. In language models, this can include instructions, user input, conversation history and other supplied content.
Explore: INTAI Governance Framework
A control is a measure designed to prevent, detect, reduce or respond to risk. AI controls can include approvals, access restrictions, testing, human review, monitoring, documentation and escalation requirements.
Explore: AI Assurance Reviews
Control effectiveness describes whether a control is appropriately designed, implemented and operating in practice to manage the risk it is intended to address.
Copilot is a name used by Microsoft for a range of AI-assisted products and capabilities designed to support people across workplace, development and other digital activities.
Explore: AI Governance Policies
Copyright and AI refers to the legal and governance questions that arise when copyrighted material is used in connection with AI systems or when AI is used to generate, transform or reproduce content. Applicable rights and obligations depend on the circumstances and relevant jurisdiction.
Data anonymisation is the process of altering information so that individuals can no longer be identified by reasonably available means. Effective anonymisation is different from pseudonymisation, where re-identification remains possible using additional information.
Explore: AI Monitoring Services
Data drift occurs when the characteristics or distribution of data encountered by an AI system change over time compared with the data used to develop, train or validate it, potentially affecting performance.
Explore: AI Data Readiness Assessment
Data governance is the system of responsibilities, standards and controls used to manage data throughout its lifecycle, including its ownership, quality, access, security, use, retention and disposal.
Explore: AI Data Readiness Assessment
Data lineage records how data moves through systems and processes, including where it originated, how it was transformed and where it has been used.
Explore: AI Governance Policies
Data minimisation is the principle that organisations should collect, use and retain only the personal data that is adequate, relevant and necessary for the defined purpose.
Data poisoning is the deliberate manipulation or contamination of data used by an AI or machine learning system with the aim of altering its behaviour, reducing its performance or creating hidden vulnerabilities.
Explore: AI Governance Policies
Data privacy concerns how personal information is collected, used, shared, retained and protected, including the rights and expectations of individuals whose information may be processed by AI systems.
A data protection impact assessment, or DPIA, is a structured process for identifying and reducing privacy risks associated with processing personal data, particularly where processing is likely to create high risks to individuals.
Explore: AI Data Readiness Assessment
Data provenance is information about where data originated, how it was collected, how it has changed and how it has been used, helping organisations assess its suitability, reliability and governance.
Explore: AI Data Readiness Assessment
Data quality describes whether data is sufficiently accurate, complete, consistent, relevant, timely and reliable for its intended purpose. Poor-quality data can materially affect AI performance and outcomes.
Explore: AI Data Readiness Assessment
Data readiness describes whether an organisation has data of sufficient quality, accessibility, governance and suitability to support a proposed AI initiative or capability.
A dataset is an organised collection of information used for analysis or computational processing. In machine learning, datasets may be used to train, validate, test or monitor models.
Explore: Governance Operating Models
Decision rights define which individuals, roles or governance bodies have authority to make, approve, challenge or escalate particular decisions.
Deep learning is a branch of machine learning that uses neural networks with multiple computational layers to learn complex patterns from large amounts of data.
Explore: AI Training · Responsible AI
A deepfake is synthetic or manipulated media created using AI to convincingly imitate or alter a person's appearance, voice or actions, potentially making fabricated content appear authentic.
Explore: EU AI Act
Under the EU AI Act, a deployer is generally a person or organisation using an AI system under its authority in a professional context, subject to the specific definitions and exclusions set out in the legislation.
Explore: AI Lifecycle
Deployment is the stage at which an AI system is introduced into a live environment where it can be used within real organisational processes, products or services.
Explore: AI Monitoring Services
Drift is a change over time in the data, environment or relationships on which an AI model depends, potentially causing its behaviour or performance to deteriorate or become less appropriate.
An embedding is a numerical representation of information such as words, documents or images that captures aspects of meaning or similarity in a form that AI systems can process and compare.
Emergent behaviour describes capabilities or patterns of behaviour that arise from a complex AI system and were not explicitly programmed or straightforwardly predicted from its individual components.
Explore: EU AI Act · AI Compliance Readiness
The EU AI Act is the European Union's legal framework for artificial intelligence. It establishes obligations based on the type and risk of AI systems and includes requirements affecting providers, deployers and other participants in the AI value chain.
An evaluation dataset is a collection of data used to assess how well an AI or machine learning model performs against defined tasks, measures or requirements.
Explore: Explainability Reviews · Responsible AI
Explainability is the extent to which the reasons, factors or processes behind an AI system's output can be communicated in a way that is meaningful to the people who need to understand or challenge it.
Explore: Bias & Fairness Audits · Responsible AI
Fairness in AI concerns whether systems and their outcomes treat people appropriately and avoid unjustified disadvantage or discrimination. What constitutes fairness can depend on the context, affected groups and applicable legal requirements.
Explore: Prompt Engineering Assessment
Few-shot prompting is a technique in which a small number of examples are included in a prompt to demonstrate the type, style or structure of response expected from an AI model.
Fine-tuning is the process of further training an existing AI model using additional data so that its behaviour or performance becomes better suited to a particular task, domain or set of requirements.
Explore: Model Governance
A foundation model is a broadly capable AI model trained on large amounts of data that can be adapted or used as the basis for a wide range of downstream applications and tasks.
Generalisation is a model's ability to perform effectively on new information or situations that were not part of the data used to train it.
Explore: EU AI Act · Model Governance
General-purpose AI refers to AI models with broad capabilities that can perform a wide range of distinct tasks and be integrated into different systems or applications. The EU AI Act establishes specific obligations relating to general-purpose AI models.
Explore: Generative AI Use Policy · AI Literacy Assessment
Generative AI is a category of artificial intelligence designed to create new content, including text, images, audio, video and software code, based on patterns learned from data and the instructions or other inputs it receives.
Explore: AI Governance Strategy · AI Governance Policies
Generative AI governance is the system of policies, responsibilities and controls used to manage how generative AI tools and models are selected, accessed, used and monitored within an organisation.
Explore: AI Governance Policies
A generative AI policy defines how people within an organisation may use generative AI, including requirements relating to approved tools, sensitive information, output review, intellectual property, accountability and prohibited uses.
Explore: AI Risk Assessment · AI Governance Policies
Generative AI risk refers to potential adverse outcomes arising from generative AI, including inaccurate content, inappropriate disclosure of information, bias, intellectual property concerns, security threats and over-reliance on generated outputs.
Explore: AI Governance Strategy
Governance is the system through which an organisation directs activity, assigns responsibilities, makes decisions, manages risk and provides oversight and accountability.
Explore: Governance Operating Models
A governance operating model defines how governance works in practice, including the roles, committees, decision rights, processes, reporting lines and escalation routes used to oversee AI across an organisation.
Explore: EU AI Act
GPAI model is shorthand for a general-purpose AI model capable of performing a broad range of tasks and being integrated into numerous downstream systems or applications.
Grounding is the practice of connecting an AI system's responses to defined information, evidence or context so that its outputs are more relevant to the task and can be more readily checked against trusted sources.
Explore: INTAI Governance Framework
An AI guardrail is a technical, procedural or organisational measure designed to constrain an AI system's behaviour or reduce the likelihood of unsafe, inappropriate or unauthorised outcomes.
Explore: AI Training · Responsible AI
An AI hallucination occurs when a generative AI system produces information that appears plausible but is inaccurate, unsupported or fabricated. Important outputs should therefore be checked rather than assumed to be correct.
Explore: AI Risk Classification
High-impact AI is a general governance term for AI whose decisions, recommendations or failures could have significant consequences for people, organisations, services, finances, safety or rights and therefore warrant stronger oversight.
Explore: EU AI Act · AI Risk Classification
A high-risk AI system is an AI system that falls within the high-risk categories defined by the EU AI Act and is therefore subject to specific legal requirements relating to matters such as risk management, data, documentation, human oversight, accuracy and monitoring.
Explore: Responsible AI
Human-centred AI is an approach to designing and using artificial intelligence that considers human needs, capabilities, rights and wellbeing throughout the system's lifecycle.
Explore: Responsible AI · AI Governance Policies
Human-in-the-loop describes a process in which a person actively participates in an AI-enabled workflow, such as by reviewing, approving, correcting or overriding an AI-generated output or decision.
Explore: Responsible AI
Human-on-the-loop describes an arrangement in which an AI system operates with a degree of autonomy while a person supervises its activity and retains the ability to intervene.
Explore: Responsible AI · AI Governance Policies
Human oversight is the set of arrangements that enable appropriately authorised people to understand, supervise, challenge, intervene in or stop an AI system when necessary.
Explore: AI Risk Assessment
An impact assessment is a structured process used to understand the potential positive and negative consequences of a proposed activity, system or decision before or during implementation.
Inference is the process through which a trained AI or machine learning model uses new input to generate an output, such as a prediction, classification, recommendation or response.
Explore: AI Risk Assessment
Information security is the protection of information and systems against unauthorised access, disclosure, alteration, destruction or disruption, commonly focused on preserving confidentiality, integrity and availability.
Explore: AI Technology Implementation
Intelligent automation combines automation with AI capabilities so that workflows can interpret information, adapt to different inputs or support decisions rather than following only fixed rules.
Explore: Explainability Reviews
Interpretability is the extent to which people can understand how an AI model works or how particular inputs and internal relationships influence its outputs.
Explore: ISO/IEC 42001 · AI Compliance Readiness
ISO/IEC 42001 is an international management system standard for artificial intelligence. It provides requirements for establishing, implementing, maintaining and continually improving an AI management system within an organisation.
Explore: AI Agents Readiness Assessment
A kill switch is a mechanism that allows an AI system, automated process or agent to be rapidly stopped, disabled or isolated when continued operation could create unacceptable risk.
A knowledge graph is a structured representation of entities and the relationships between them, enabling systems to connect and reason across linked information.
A language model is an AI model designed to process patterns in language and predict or generate sequences of text or other language-based outputs.
Explore: Model Governance · AI Training
A large language model, or LLM, is an AI model trained on large amounts of text and other data to process and generate language and, in many modern systems, work across additional types of information.
LLM stands for large language model, a type of AI model designed to process and generate language. LLMs underpin many modern generative AI assistants and applications.
Explore: AI Training
Machine learning is a field of artificial intelligence in which computational models learn patterns from data so they can make predictions, classifications or other outputs without every decision rule being explicitly programmed.
Explore: Model Governance
A machine learning model is a computational model developed by learning patterns from data and then applying those learned relationships to new information.
Explore: AI Lifecycle · AI Monitoring Services
Machine learning operations, commonly known as MLOps, is the set of practices used to develop, deploy, operate, monitor and maintain machine learning models reliably throughout their operational lifecycle.
Explore: AI Risk Classification
A material AI system is an AI system considered sufficiently significant to an organisation's operations, decisions, customers, finances, reputation or risk profile to warrant formal governance and senior oversight.
Microsoft 365 Copilot is Microsoft's AI assistant integrated with Microsoft 365 applications and services. It can work with organisational information that the individual user is permitted to access, making information governance and permissions important readiness considerations.
MLOps, short for machine learning operations, refers to the practices, processes and technologies used to move machine learning models from development into reliable production operation and ongoing monitoring.
Explore: Model Governance
In AI and machine learning, a model is a computational representation that processes inputs to produce outputs such as predictions, classifications, recommendations or generated content.
Explore: Model Governance
A model card is a structured document describing an AI or machine learning model, commonly including its intended purpose, performance, limitations, evaluation approach and important considerations for responsible use.
Explore: AI Monitoring Services
Model drift describes deterioration or change in a model's performance or behaviour over time as the environment, underlying relationships or data it encounters change.
Explore: AI Assurance Reviews
Model evaluation is the structured assessment of how well an AI model performs against defined measures, tasks, requirements and relevant risk considerations.
Explore: Model Governance · AI Assurance Reviews
Model governance is the system of responsibilities, standards, controls and oversight used to manage models throughout their lifecycle, including development, validation, approval, deployment, monitoring, change and retirement.
Explore: Model Governance
A model inventory is a maintained record of models used by an organisation, typically including their purpose, ownership, status, materiality, risk classification and key lifecycle information.
Explore: AI Monitoring Services
Model monitoring is the ongoing review of a model's performance, behaviour and relevant operating conditions to identify deterioration, drift, unexpected outcomes or other issues requiring action.
Explore: AI Monitoring Services
Model performance describes how effectively a model carries out its intended task when assessed against relevant measures such as accuracy, reliability, error rates or task-specific quality criteria.
Explore: Model Governance · AI Assurance Reviews
Model risk is the possibility of adverse consequences arising from incorrect, poorly designed, inadequately understood or inappropriately used models and their outputs.
Explore: AI Assurance Reviews
Model robustness is the ability of an AI model to continue performing appropriately when faced with changing conditions, unexpected inputs or reasonable variations in its operating environment.
Explore: AI Assurance Reviews
Model testing is the systematic examination of an AI model's behaviour and performance under defined conditions before and, where appropriate, after deployment.
Model training is the process through which an AI or machine learning model learns patterns from data by adjusting its internal parameters.
Explore: Model Governance · AI Assurance Reviews
Model validation is the structured evaluation of whether a model is appropriately designed and performs sufficiently for its intended use, including review of its methodology, assumptions, data, performance and limitations.
Explore: AI Monitoring Services
Monitoring is the ongoing observation and review of a system, process or control to identify changes, deterioration, unusual behaviour or emerging risks.
Multimodal AI refers to AI systems capable of processing or generating more than one type of information, such as text, images, audio and video, within the same model or application.
Natural language processing, or NLP, is a field of AI concerned with enabling computer systems to process, analyse and generate human language in written or spoken form.
A neural network is a machine learning architecture made up of interconnected computational units arranged in layers and designed to learn patterns and relationships from data.
Explore: NIST AI RMF
The NIST AI Risk Management Framework is a voluntary framework developed by the US National Institute of Standards and Technology to help organisations manage AI risks through the functions Govern, Map, Measure and Manage.
NLP stands for natural language processing, the field of AI concerned with enabling computer systems to work with human language, including text and speech.
Explore: AI Vendor Assessments
Open-source AI generally refers to AI software or models made available under terms that permit defined forms of access, use, modification or redistribution. The exact level of openness varies and should be assessed from the applicable licence and released components.
An open-weight model is an AI model whose trained parameter weights are made available for others to access or run, although this does not necessarily mean that all training data, source code or development details are open.
Explore: Responsible AI
Output validation is the process of checking an AI-generated output for accuracy, quality, appropriateness and compliance with relevant requirements before it is relied upon or used for a consequential purpose.
Overfitting occurs when a machine learning model learns its training data too closely and performs well on that data but does not generalise effectively to new or unseen information.
Personal data is information relating to an identified or identifiable person and is subject to applicable data protection requirements when processed through AI systems.
Explore: EU AI Act
Post-market monitoring is the systematic collection and review of information about an AI system after it has been placed on the market or put into service so that emerging issues can be identified and addressed.
Predictive AI uses data and computational models to estimate future events, outcomes or probabilities, such as demand, customer behaviour, equipment failure or financial risk.
A predictive model uses patterns in available data to estimate future events, values, behaviours or probabilities.
Privacy by design is the principle of considering and embedding appropriate data protection safeguards throughout the design and operation of systems and processes rather than addressing privacy only after deployment.
Explore: EU AI Act
A prohibited AI practice is a use of artificial intelligence that is not permitted under applicable law. The EU AI Act identifies specific AI practices that are prohibited within its scope.
Explore: Prompt Engineering Assessment
A prompt is an instruction, question or other input provided to a generative AI system to guide the response or output it produces.
Explore: Prompt Engineering Assessment · AI Training
Prompt engineering is the practice of designing and refining instructions, context, examples and constraints to help a generative AI system produce outputs that are more useful, consistent and appropriate for the intended task.
Explore: AI Risk Assessment · AI Agents Readiness Assessment
Prompt injection is an attack or manipulation technique in which instructions are introduced into an AI system's inputs in an attempt to override intended instructions, alter its behaviour or cause unauthorised actions or disclosure.
Prompt leakage occurs when instructions, context or information intended to remain hidden within an AI system are unintentionally revealed through its outputs or interactions.
Explore: Prompt Engineering Assessment
A prompt library is an organised collection of reusable prompts designed to support more consistent, efficient and effective use of generative AI for recurring tasks.
Explore: Prompt Engineering Assessment
Prompt management is the structured process of developing, testing, approving, maintaining and version-controlling prompts used in repeatable or operational AI workflows.
Explore: Prompt Engineering Assessment
A prompt template is a reusable structure for giving an AI system instructions while allowing specific information, context or requirements to be changed for each use.
Explore: EU AI Act
Under the EU AI Act, a provider is generally a person or organisation that develops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts it into service under its own name or trademark.
RAG stands for retrieval-augmented generation, an approach that retrieves relevant information from external sources and provides it to a generative AI model as additional context when producing a response.
A reasoning model is an AI model designed or optimised to perform more complex analysis, planning and multi-step problem-solving before producing a response or taking an action.
Explore: AI Assurance Reviews
Red teaming is a structured form of adversarial testing in which testers deliberately challenge a system to identify vulnerabilities, unsafe behaviours, control weaknesses and potential misuse scenarios.
A regression model is a statistical or machine learning model used to estimate numerical values or relationships, such as predicting costs, demand, prices or other continuous outcomes.
Reinforcement learning is a machine learning approach in which a system learns behaviour by taking actions within an environment and receiving signals that reward or penalise different outcomes.
Explore: Responsible AI · INTAI Governance Framework
Responsible AI is the approach of developing and using artificial intelligence in ways that appropriately address matters such as fairness, accountability, transparency, safety, privacy, human oversight and wider impacts.
Explore: Responsible AI
Responsible AI principles are the high-level commitments used to guide how an organisation develops and uses AI, commonly addressing areas such as fairness, accountability, transparency, privacy, safety and human oversight.
Retrieval-augmented generation, or RAG, combines information retrieval with generative AI by finding relevant information from external sources and supplying it as context to a model when producing a response.
Explore: AI Risk Assessment · AI Governance Strategy
Risk appetite is the amount and type of risk an organisation is willing to accept in pursuit of its objectives. For AI, it can help determine which uses require stronger controls, senior approval or should not proceed.
Explore: AI Risk Assessment · AI Risk Assessment Tool
A risk assessment is a structured process for identifying, analysing and evaluating potential risks so that informed decisions can be made and appropriate controls established.
Explore: AI Risk Classification
A risk-based approach applies governance, assurance and controls in proportion to the potential impact and risk of an AI system rather than treating every AI use in the same way.
Explore: AI Risk Classification
Risk classification is the process of grouping activities, systems or use cases into defined risk categories so that governance, approval, assurance and control requirements can be applied proportionately.
Explore: AI Risk Assessment
Risk management is the coordinated process of identifying, assessing, treating, monitoring and communicating risks so that informed decisions can be made.
Explore: INTAI Templates
A risk register is a maintained record of identified risks together with information such as their severity, controls, treatment actions, accountable owners and current status.
Explore: AI Assurance Reviews
A safety case is a structured argument supported by evidence explaining why a system is considered acceptably safe for a defined purpose and operating environment.
Explore: AI Governance Policies
Sensitive data is information requiring additional protection because unauthorised access, disclosure, alteration or misuse could cause significant harm to individuals or organisations.
Explore: AI Governance Strategy · AI Governance Policies
Shadow AI is the use of AI tools, models or services within an organisation without appropriate approval, visibility or governance, often creating unmanaged risks relating to data, security, quality and compliance.
Supervised learning is a machine learning approach in which a model learns from examples containing known or labelled outcomes and uses those patterns to make predictions about new data.
Explore: AI Governance Policies · EU AI Act
Synthetic content disclosure is the practice or requirement of informing people when content has been generated or materially manipulated using artificial intelligence where disclosure is appropriate or legally required.
Synthetic data is artificially generated information designed to reproduce selected characteristics of real-world data and may be used for purposes such as model development, testing, simulation or privacy-preserving analysis.
Explore: Responsible AI · AI Governance Policies
Synthetic media is content generated or materially altered using artificial intelligence, including images, video, audio and other media that may imitate or create realistic people, objects, events or environments.
Explore: Model Governance · AI Assurance Reviews
A system card is a structured document describing an AI system's capabilities, intended uses, limitations, evaluation results and relevant safety, risk or governance considerations.
Explore: Prompt Engineering Assessment
A system prompt is a set of higher-level instructions used to guide the role, behaviour and operating boundaries of a generative AI model or application.
Explore: EU AI Act · AI Risk Assessment
Systemic risk is the possibility that an AI system or model could create significant adverse effects at broad scale because of its capabilities, reach, interconnectedness or widespread use.
Temperature is a model setting used in some generative AI systems to influence the variability of generated outputs, with higher values generally producing more varied responses and lower values tending to produce more consistent outputs.
Explore: AI Vendor Assessments · AI Vendor Assessment Tool
Third-party AI is an AI model, system, service or capability supplied, hosted or materially controlled by an external organisation rather than developed and operated entirely in-house.
Explore: AI Training
A token is a unit of information processed by a language model. In text, a token may represent a whole word, part of a word, punctuation or another character sequence depending on how the model processes language.
Explore: AI Agents Readiness Assessment
Tool use is the ability of an AI system or agent to interact with external applications, APIs, databases or services to retrieve information, complete tasks or perform actions.
Explore: INTAI Governance Framework · AI Assurance Reviews
Traceability is the ability to follow and evidence important information, decisions, changes and actions throughout an AI system's lifecycle, supporting accountability, investigation, assurance and regulatory compliance.
Explore: AI Data Readiness Assessment
Training data is the information used during the development of a machine learning model so that it can learn patterns, relationships or representations relevant to its intended capabilities.
A transformer is a neural network architecture that uses attention mechanisms to process relationships within information and forms the technical foundation of many modern large language and generative AI models.
Explore: Responsible AI · AI Governance Policies
Transparency is the practice of making appropriate information available about an AI system, including its use, purpose, capabilities, limitations, responsibilities and significant impacts where relevant.
Explore: EU AI Act · AI Compliance Readiness
A transparency obligation is a legal, regulatory or organisational requirement to provide specified information about the use, operation or outputs of an AI system to relevant people, users or authorities.
Explore: Responsible AI · AI Assurance Reviews
Trustworthy AI is a broad concept describing artificial intelligence that can be relied upon appropriately because relevant considerations such as reliability, safety, fairness, accountability, transparency, privacy, security and human oversight have been addressed.
Unsupervised learning is a machine learning approach in which a model analyses data without predefined outcome labels in order to identify patterns, structures, similarities or groupings.
Explore: AI Opportunity Assessment · INTAI Templates
A use case is a defined application of technology to a particular business problem, opportunity, task or process, with a clear purpose and intended outcome.
Explore: AI Assurance Reviews · Model Governance
Validation is the structured process of gathering evidence to determine whether a model, system, control or process performs sufficiently for its intended purpose and defined requirements.
Explore: Model Governance · AI Data Readiness Assessment
Validation data is data used during model development to assess performance and support decisions such as model selection or tuning without relying on the final independent test dataset.
Explore: AI Technology Implementation
A vector database is a database designed to store and search numerical representations known as embeddings, enabling systems to retrieve information based on similarity and meaning rather than exact keyword matches alone.
Explore: AI Technology Implementation
A vector embedding is a numerical representation of information designed to capture characteristics or meaning so that similar items can be identified, compared and retrieved computationally.
Explore: AI Assurance Reviews
Verification is the process of checking whether a system, component or process has been built or implemented in accordance with specified requirements.
Explore: AI Technology Implementation
A vision model is an AI model designed to process and analyse visual information such as images or video for tasks including recognition, classification, detection and interpretation.
Explore: AI Technology Implementation
A vision-language model is an AI model designed to process and relate both visual and language-based information, enabling tasks such as image understanding, visual question answering and multimodal content analysis.
Weights are numerical parameters within a machine learning model that are adjusted during training and influence how the model processes inputs and produces outputs.
Explore: AI Technology Implementation
Workflow automation is the use of technology to execute, coordinate or move tasks and information through a defined process with reduced need for manual intervention.
Zero-shot learning is the ability of an AI model to perform a task or recognise a category for which it has not been given specific labelled training examples, using knowledge learned from related information.
Explore: Prompt Engineering Assessment
Zero-shot prompting is the practice of asking an AI model to perform a task using instructions alone without providing examples of the desired output within the prompt.
Turn insight into action with an INTAI AI Assessment .