Auditable & Trustworthy AI

AI governance is the framework of policies, controls, and accountability that an organisation puts in place to ensure its AI systems are used safely, legally, and transparently, defining who is responsible, how outputs are checked, and how risks are managed.

AI hallucination is when an AI system produces output that is fluent and confident but factually wrong or fabricated, presenting invented information, sources, or reasoning as if it were true because the model generated it from patterns rather than evidence.

AI provenance is the documented origin and history of the data, sources, and models behind an AI output, recording where information came from, how it was transformed, and which model produced the result, so its lineage can be verified.

AI traceability is the ability to follow any AI-generated output back through the exact inputs, sources, and processing steps that produced it, creating an unbroken chain from the final result to the evidence and logic that stand behind it.

AI transparency is the degree to which an AI system's data, sources, reasoning, and limitations are open to inspection, so users understand what the system is doing, what it is based on, and where its answers come from rather than treating it as an opaque tool.

Auditable AI is an artificial intelligence system whose outputs can be inspected, explained, and traced back to their sources and reasoning, so that any recommendation or result can be reviewed, verified, and defended by a human after the fact.

Black-box AI is an artificial intelligence system whose internal reasoning is hidden or inaccessible, so users see the output but cannot examine how it was reached, which inputs drove it, or whether it can be trusted for a given decision.

Deterministic AI is a system whose behaviour follows fixed, defined rules or structures so that the same inputs produce the same, predictable outputs every time, in contrast to probabilistic models whose answers can vary and cannot be guaranteed to repeat.

Explainable AI (XAI) is a set of methods and system designs that make the reasoning behind an AI model's outputs understandable to humans, so that people can see why a given prediction, recommendation, or answer was produced rather than accepting it on trust.

Grounded AI is artificial intelligence that produces outputs anchored in verified, retrievable sources rather than generating text from statistical patterns alone, so every answer rests on real evidence that can be located and confirmed.

Human-in-the-loop is an AI design pattern in which people review, guide, or approve the system's outputs at key points, keeping human judgement in control of consequential decisions rather than letting the AI act autonomously without oversight.

Responsible AI is the practice of designing and using artificial intelligence in ways that are fair, transparent, accountable, and safe, ensuring systems respect human oversight, avoid harm, and produce outputs that can be explained and stood behind.

Retrieval-augmented generation (RAG) is an AI technique that retrieves relevant information from a trusted knowledge source and supplies it to a language model at the moment of generation, so the model's output is grounded in real, current evidence rather than memory alone.

Source attribution is the practice of linking each part of an AI output to the specific document, dataset, or reference it was drawn from, so a reader can see exactly where a claim came from and open the underlying evidence to verify it.