Appendix E. Glossary

AI agent: A system that carries out multi-step tasks toward a goal, sometimes using tools or external services.

Algorithm: A defined procedure for processing inputs or solving a problem.

Automation bias: The tendency to over-trust an automated recommendation or decision.

Bias: A systematic pattern that can distort outcomes or disadvantage people or groups.

Data minimisation: Limiting collection and use of information to what is necessary for a stated purpose.

Deepfake: Synthetic or manipulated audio, image, or video that can make a person or event appear authentic when it is not.

Distribution shift: A change between the conditions represented in development data and those encountered in use.

Generative AI: AI designed to create new content such as text, images, audio, video, or code.

Hallucination / confabulation: An AI-generated claim that is false, fabricated, or unsupported by available evidence.

Human oversight: Meaningful ability for a responsible person to understand, challenge, override, or stop a system.

Large language model (LLM): A model trained on large amounts of text and related data to process or generate language.

Model drift: A decline or change in performance as data, users, or conditions evolve.

Prompt injection: An attempt to manipulate an AI system through instructions embedded in untrusted input or retrieved content.

Provenance: Information about the origin, creation, modification, or handling of content or data.

Risk register: A structured record of risks, impacts, controls, owners, and review status.

Synthetic media: Content created or materially altered by computational generation or manipulation.

Vendor lock-in: Dependence on a supplier that makes switching systems or moving data difficult.

Verification: Checking a claim, output, or decision against reliable evidence or an independent process.