• workshop_dense_100#slide 16

In RAIDT, run-level governance means governing one configured generative AI use in context through evidence that can later be inspected, reconstructed, and scored. A run is not merely a model invocation in the narrow technical sense; it is a configured organisational use event shaped by task context, prompts, retrieval, tools, settings, and human oversight. RAIDT therefore treats run as the unit of governance and requires a run-level evidence pack that records the use purpose, stakes, provenance, outputs, and review decisions for that event.

The importance of this approach is twofold. Conceptually, it provides the missing evidentiary layer between broad governance principles and concrete organisational action. Practically, it allows reviewers to assess governance readiness through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) and to express the result as a score profile with anchors 1=missing / 3=partial / 5=audit-ready. The papers also stress influence methods as governance interventions: structured prompting, retrieval augmentation, PEFT or LoRA, preference-based alignment, and review controls are all configuration choices that change governance outcomes and therefore must be recorded. Run-level governance matters because it makes those choices visible, lets organisations compare configurations across repeated uses, and supports challenge, adjudication, and improvement when a particular output becomes consequential or disputed.

A cybersecurity team uses a GenAI assistant for alert triage. One configuration relies on a simple prompt; another adds structured prompting, retrieval of recent incident knowledge, and explicit reviewer sign-off. RAIDT would treat those configurations as different governed runs, not as the same system in generic form.

For one alert, the run-level evidence pack would record the prompt template, retrieved artefacts, model version, parameters, generated recommendation, and analyst decision. Reviewers could then examine whether the recommendation was dependable across repeat runs and whether the supporting evidence made the run auditable and traceable. That concrete, inspectable treatment is why run-level governance matters in RAIDT.

• 08-RAIDT_Foundations_M_V50
• 16-RAIDT-Audit-Accountability_M_v05