Legal and technical concepts
AI Glossary
Clear explanations of the technologies, organisations and legal concepts shaping the law of AI.
Directory
All terms
75 definitions
Agentic AI
AI systems that pursue goals through autonomous planning and action, with limited human intervention.
Core AIDataRiskTechnicalRead definitionAI Architecture
The overall design of an AI system, including its models, data pipelines, and compute. Affects performance, compliance, and risk.
Core AIModelsTechnicalLegalRead definitionAI Explainability
The ability to understand how an AI system makes decisions. Crucial for trust, compliance, and legal accountability.
TransparencyLegalTechnicalGovernanceRead definitionAlgorithm
A set of rules or instructions given to an AI system to help it learn and make decisions.
Core AITechnicalRead definitionAlignment
The process of ensuring that an artificial intelligence system's goals and behaviors closely match human values and intentions.
EthicsRiskRead definitionAlignment Drift
When an AI system gradually deviates from its intended goals or safety constraints over time or usage.
EthicsLegalRiskRead definitionAnthropic
An AI safety company founded in 2021, known for developing the Claude family of large language models with a focus on AI safety and alignment research.
BusinessIndustryRead definitionArtificial General Intelligence (AGI)
Highly autonomous systems that outperform humans at most economically valuable work.
Core AIModelsRead definitionArtificial Intelligence (AI)
Computer science focused on creating systems that can perform tasks typically requiring human intelligence.
Core AIDataLegalRead definitionBias in AI
Systematic errors in AI systems that can lead to unfair or discriminatory outcomes.
Core AIDataEthicsRiskRead definitionChatGPT
A large language model developed by OpenAI, designed to generate human-like text responses in conversational settings.
BusinessModelsRead definitionClaude
A large language model developed by Anthropic, designed with a focus on safety, transparency, and usability.
BusinessModelsRead definitionComputer Vision
The field of AI focused on enabling machines to interpret and understand visual information from the world.
Core AITechnicalRead definitionConstitutional AI
A training methodology where models follow principles or "constitutional rules" during reinforcement learning.
Core AIEthicsGovernanceModelsRead definitionContext Window
The span of text an LLM can "remember" or consider in a single interaction or prompt.
ModelsTechnicalRead definitionData Catalog
A structured inventory of an organization's data assets, including metadata and lineage.
DataTechnicalRead definitionData Ingestion
The process of collecting and importing data for immediate use or storage.
DataTechnicalRead definitionData Labeling
The act of tagging data with meaningful labels for training supervised machine learning models.
DataTechnicalRead definitionData Lake
A centralised repository that stores raw structured and unstructured data at scale.
DataLegalTechnicalRead definitionData Pipeline
A series of processes that automate data movement and transformation from source to destination.
DataTechnicalRead definitionData Sanitisation
The process of cleaning data by removing or masking sensitive or inaccurate information.
DataRiskSecurityRead definitionData Validation
Ensuring that data meets quality and formatting standards before further use.
DataRiskTechnicalRead definitionData Warehouse
A repository for storing curated, structured data optimized for reporting and analytics.
DataTechnicalRead definitionDataset
A collection of data points, often used to train or evaluate AI models.
DataRead definitionDeep Learning
A subset of machine learning using deep neural networks with many layers to extract features from data.
Core AIDataTechnicalRead definitionFine-tuning
The process of adapting a pre-trained AI model to perform specific tasks by training it on additional, domain-specific data while preserving its foundational capabilities.
DataGovernanceLegalModelsTechnicalRead definitionFoundation Model
A large-scale model trained on massive data that serves as the base for adaptation to a wide range of tasks.
DataLegalModelsTechnicalRead definitionGemini
A family of multimodal AI models developed by Google DeepMind, integrating text, code, images, and more.
BusinessModelsRead definitionGenerative AI
AI systems that can generate new content such as text, images, music, or code.
Core AIModelsRead definitionGenerative Pre-trained Transformer (GPT)
A type of language model that uses transformer architecture and is pre-trained on large text corpora.
ModelsTechnicalRead definitionGitHub Copilot
An AI-powered coding assistant developed by GitHub that suggests code, functions, and solutions from natural-language prompts.
BusinessModelsTechnicalRead definitionHallucinations
When AI generates plausible-sounding but incorrect or fabricated content not grounded in the input data.
DataModelsRiskRead definitionHugging Face
An open-source AI platform and community offering tools, models, and datasets for building NLP and ML systems.
BusinessDataLegalModelsRead definitionHuman-in-the-Loop (HitL)
A model workflow where human oversight is included to guide or validate AI decisions.
EthicsGovernanceRead definitionHybrid Search
Combines traditional keyword search with AI-powered semantic search to improve information retrieval.
DataLegalTechnicalRead definitionInference
The process of applying a trained AI model to new data to generate outputs or predictions.
DataModelsTechnicalRead definitionLarge Language Model (LLM)
A deep learning model trained on vast text corpora capable of understanding and generating human-like language.
Core AIModelsRead definitionLLaMA (Large Language Model Meta AI)
A family of open-weight large language models developed by Meta (Facebook) to advance accessible AI research.
BusinessCore AIModelsRead definitionLLM Bias
The tendency of large language models to reflect or amplify societal or dataset-based biases.
DataEthicsModelsRiskRead definitionLLM Connector
Middleware that integrates LLMs into existing systems, allowing seamless input/output between applications and models.
ModelsTechnicalRead definitionLLM Hyperparameters
Settings that govern how a model is trained or generates responses, such as learning rate or temperature.
ModelsTechnicalRead definitionLLM Jailbreaking
A technique to bypass safeguards in language models, often to trigger forbidden or unsafe outputs.
LegalModelsRiskSecurityRead definitionLLM Quality Assurance
Processes to evaluate, monitor, and ensure the performance and safety of deployed LLMs.
GovernanceLegalModelsTechnicalRead definitionLLM Settings
Configuration options that influence how a model behaves in use (e.g., max tokens, temperature, top_p).
LegalModelsTechnicalRead definitionLLMOps
The discipline of managing the lifecycle of LLMs — including deployment, scaling, versioning, monitoring, and governance.
GovernanceLegalModelsTechnicalRead definitionMachine Learning
A subset of artificial intelligence that enables systems to automatically learn and improve performance on tasks through experience without being explicitly programmed for each specific outcome.
Core AITechnicalLegalRiskRead definitionMeta
The parent company of Facebook, Instagram, and WhatsApp. A major AI developer behind the LLaMA family of open-weight large language models.
BusinessIndustryRead definitionMetadata
Data about data. Helps classify, search, and manage information more effectively.
DataLegalTechnicalRead definitionMicrosoft Copilot
A family of AI assistants developed by Microsoft and integrated into products such as Windows, Microsoft 365, and Bing.
BusinessModelsTechnicalRead definitionMLOps
A set of practices to deploy, monitor, and maintain machine learning models in production reliably.
DataGovernanceTechnicalRead definitionModel Collapse
The idea that AI models trained on data generated by other models may gradually degrade in quality, novelty, or accuracy.
DataModelsRiskRead definitionRecipe
In the context of AI and data processing, a set of instructions or steps for performing a specific task or achieving a particular outcome.
DataTechnicalRead definitionRed Team Testing
A structured security assessment process where teams systematically probe AI systems to identify potential vulnerabilities, harmful outputs, and safety failures before deployment. Red teaming involves adversarial testing across multiple risk categories including bias, toxicity, jailbreaking attempts, and misuse scenarios. The practice helps developers understand edge cases and implement safeguards. Anthropic maintains a Bug Bounty Program and publishes research on red teaming methodologies.
GovernanceRiskSecurityRead definitionReinforcement Learning
A type of machine learning where an agent learns to make decisions by taking actions in an environment to maximise a reward.
Core AITechnicalRead definitionRetrieval-Augmented Generation (RAG)
A method where LLMs dynamically retrieve external documents or facts to enhance response accuracy.
DataLegalModelsTechnicalRead definitionShadow Library
An unauthorised digital repository that aggregates and distributes copyrighted books, academic papers, and other works without the permission of rights holders.
EthicsDataRiskTechnicalRead definitionStructured Data
Data organised in a defined format, such as tables or spreadsheets, making it easily searchable and analyzable.
DataTechnicalRead definitionStructured vs. Unstructured Data
Structured data fits neatly into tables; unstructured data is free-form like text, audio, or video.
DataRead definitionSuperintelligent AI
A hypothetical form of AI that surpasses human intelligence across all domains.
Core AIEthicsRiskRead definitionSupervised Learning
A machine learning approach that uses labeled data to train a model to make predictions or classifications.
Core AIDataTechnicalRead definitionSynthetic Data
Artificially generated data used to train, augment, or fine-tune AI models, often to address data scarcity or privacy concerns.
DataLegalRiskSecurityRead definitionSystem Prompt
A hidden instruction given to the LLM to steer its behavior during interactions.
LegalModelsTechnicalRead definitionTemperature (LLM)
A setting that controls the randomness and creativity of AI output. Higher = more creative, lower = more focused.
ModelsTechnicalRead definitionThird-Party Service LLM
A language model hosted and offered by an external vendor via API.
BusinessModelsRead definitionToken
The smallest meaningful unit of text processed by an AI language model, such as a word, sub-word, or punctuation mark.
ModelsTechnicalRead definitionTokenisation
The process of breaking down text into chunks ("tokens") that are used as inputs by LLMs.
ModelsTechnicalRead definitionTopic Modeling
An unsupervised learning technique for identifying abstract themes in large text collections.
DataLegalModelsTechnicalRead definitionTraceability
The ability to track how and why an AI system made a specific decision or output.
EthicsGovernanceLegalRead definitionTraining Data
The dataset used to teach a model to understand patterns and make predictions.
DataLegalTechnicalRead definitionTransformer
A neural network architecture that revolutionised NLP by enabling parallel processing and contextual awareness.
ModelsTechnicalRead definitionTuring Test
A test proposed by Alan Turing to determine if a machine exhibits human-like intelligence by assessing whether a human can distinguish it from another human in conversation.
Core AIEthicsRead definitionUnsupervised Learning
A machine learning method that uncovers patterns in data without labeled outcomes.
Core AIDataTechnicalRead definitionVibe Coding
A programming style powered by LLMs, where developers rely on conversational prompting and intuition over syntax.
BusinessTechnicalRead definitionWatermarking
Embedding detectable signals in AI-generated content to verify authenticity or origin.
LegalRiskSecurityTechnicalRead definitionZombie Internet
The concept that a growing share of online content is AI-generated, creating a web that feels alive but lacks real human interaction.
EthicsIndustryRiskRead definition