AI glossary
The terms, in plain English.
The AI governance, EU AI Act and production-AI terms we work with, defined without the jargon. If a definition here is wrong or out of date, tell us.
- Governed agentic AI
- An AI agent or automation that acts on a goal, but does so within boundaries you control: bounded permissions, observable decisions, evaluation that proves it works, and the ability to reverse or override it. The opposite of autonomy you hope about. AI agents and automation →
- AI agent
- A system that takes a goal and completes it by planning steps, making decisions and using tools, rather than waiting for a human to drive each action. Distinct from a chatbot, which responds but does not act. AI agents and automation →
- AI automation
- Using AI to remove repetitive, multi-step work end to end, including the steps that need a little judgment and so resist traditional rule-based automation. AI agents and automation →
- AI governance
- The policies, controls, oversight and evidence that keep an organisation’s AI use safe, explainable and defensible. Done well it is architecture decided up front, not paperwork assembled before an audit. AI governance →
- AI Management System (AIMS)
- A working management system for AI: the policies, controls, risk processes, records and review cycles an organisation actually runs to govern its AI. It is the core of what ISO/IEC 42001 asks you to build. AI governance →
- ISO/IEC 42001
- The international standard for AI management systems. It sets out how to govern AI across its lifecycle, from policy and risk assessment to controls, documentation and continual improvement, and you can be certified against it. AI governance readiness check →
- EU AI Act
- The European Union’s regulation of AI, which classifies AI systems by risk and places obligations accordingly. It applies extraterritorially: a non-EU organisation is in scope if its AI’s outputs affect people in the EU. Penalties reach €35M or 7% of global turnover. EU AI Act readiness check →
- Deployer (EU AI Act)
- An organisation that uses an AI system under its own authority, rather than building it and placing it on the market. Most businesses are deployers. Deployer duties for high-risk AI (Articles 26 and 27) and transparency (Article 50) apply from 2 August 2026. EU AI Act readiness check →
- Provider (EU AI Act)
- An organisation that develops an AI system, or has one developed, and puts it on the market or into service under its own name or brand. Providers carry a heavier set of obligations than deployers. You can become a provider by substantially modifying or rebranding someone else’s AI system (Article 25). EU AI Act readiness check →
- High-risk AI system
- Under the EU AI Act, an AI system used in a context the regulation treats as high-risk, such as employment, credit, education or essential services. High-risk use triggers the strictest deployer obligations, including human oversight and, where required, a fundamental-rights impact assessment. EU AI Act readiness check →
- Fundamental Rights Impact Assessment (FRIA)
- An assessment certain deployers must carry out before putting a high-risk AI system into use, identifying the risks it poses to people’s fundamental rights and how those risks will be managed. It connects to, but is not the same as, a data-protection impact assessment (DPIA). AI governance →
- AI literacy (Article 4)
- The EU AI Act duty, in force since February 2025, to ensure the people who use or are affected by your AI have sufficient understanding to do so responsibly. It applies regardless of risk tier and is one of the first obligations to bite. AI governance →
- Prohibited AI practices (Article 5)
- AI uses banned outright under the EU AI Act, already in force, including social scoring, certain biometric categorisation, and emotion recognition in the workplace. Using a prohibited system carries the heaviest penalties. EU AI Act readiness check →
- Retrieval-augmented generation (RAG)
- A technique that grounds an AI model’s answers in your own data by retrieving relevant documents at query time and giving them to the model, rather than relying on what the model memorised. It is how you get answers based on your knowledge, not guesses. Custom software →
- Model Context Protocol (MCP)
- An open standard for connecting AI models and agents to external tools, data and systems through a consistent interface, so an AI can act inside your real systems rather than in isolation. AI agents and automation →
- Evaluation harness (evals)
- A repeatable test suite that measures whether an AI system does the right thing across many realistic cases, turning "we think it’s better" into a number. It is what separates a system you can demo from one you can ship and defend. Why evals come first →
- Human-in-the-loop
- A design where a person can meaningfully review, approve or override what an AI system does at the points that matter, so autonomy stays under human control rather than acting unchecked. AI agents and automation →
- Production AI
- AI that runs in front of real users and real consequences, not a demo. Production-grade means tested, monitored, secured, documented and governed, so it keeps working and stays accountable over time. AI agents and automation →