• integrated_82#Q1.8

In RAIDT, governance means the organisational capacity to make generative AI use controllable, reviewable, and answerable in practice rather than merely acceptable in principle. The governance papers frame this in classic Information Systems terms: decision rights, accountability arrangements, oversight routines, documentation practices, and socio-technical coordination that connect digital operation to organisational aims. RAIDT adopts that logic, but argues that generative AI creates a more granular problem because behaviour is materially shaped at run time through prompts, retrieved context, tools, settings, and human review actions. For that reason, RAIDT treats the run as the unit of governance.

This is why RAIDT is defined as a run-level evidence framework for governance of generative AI in organisational work. Its practical outputs are a run-level evidence pack and a score profile. The run-level evidence pack records what was intended, how the system was configured, what inputs and sources were used, what output was produced, and what checks or approvals occurred. The score profile then assesses governance readiness across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In RAIDT terminology, this anchored profile can be read as anchors 1=missing / 3=partial / 5=audit-ready. Read together, these artefacts make governance the ability to constrain use through roles and limits, evidence what happened, review and contest outputs, improve routines after failures, and account for AI-supported work in a specific organisational context.

An HR team uses a generative AI assistant to draft candidate summaries for a shortlisting meeting. Under RAIDT, governance is not satisfied by a general AI policy alone. Each shortlisting run needs a run-level evidence pack showing the approved prompt template, the job criteria used, any retrieved policy guidance, the model and settings, the human reviewer, and whether the draft was amended before being discussed.

The resulting score profile makes the run reviewable. If Responsibility and Traceability are strong but Interpretability is only partial, the team can see that it has records and approvals but weak explanation of why particular candidates were emphasised. That creates a concrete improvement task before the workflow is used again. If a candidate later challenges the process, the organisation can reconstruct what happened in that run rather than relying on vague assurances that the system was used responsibly.

• 15-RAIDT-IS-Governance_M_v07
• 13-RAIDT-Evidence-Review_M_v10