• workshop_dense_100#slide 25

In RAIDT, AI governance can be defined as the socio-technical arrangement through which an organisation directs, constrains, evidences, reviews, and accounts for generative AI use in work. It includes decision rights, accountability structures, oversight routines, documentation, monitoring, and coordination with organisational aims. The responsible-AI paper adds why this is necessary: AI systems used in uncertain managerial contexts need not only performance, but explainability, uncertainty awareness, ethical safeguards, and governance capability. The governance papers then sharpen the point by arguing that these requirements must be made inspectable in a specific configured use rather than left at the level of policy aspiration.

A simple RAIDT example is a workflow in which a generative AI system produces text or recommendations that influence records, decisions, or access to services. In such a case, governance matters because the relevant dispute will usually concern one actual use event: what was asked, what sources were retrieved, what configuration shaped the output, what review took place, and who relied on the result. RAIDT addresses that problem by treating the run as the unit of governance and by operationalising AI governance through a run-level evidence pack, a score profile, and the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). Using RAIDT terminology, the score profile can be interpreted with anchors 1=missing / 3=partial / 5=audit-ready. This matters because it turns governance from a broad promise into evidence that can support review, intervention, organisational learning, and accountable improvement.

Consider a local authority that uses generative AI to draft letters explaining whether an applicant will receive a welfare-related service. This is an AI-governance problem because the output influences access to services and may later be challenged. Under RAIDT, the organisation does not rely only on policy documents or vendor assurances. It captures a run-level evidence pack for each letter, including the prompt template, retrieved policy guidance, model settings, generated output, reviewer actions, and any escalation or override.

If the score profile shows Traceability and Auditability are strong but Responsibility is only partial, managers can see that the record is reconstructable yet role boundaries or approval rules remain weak. That finding supports immediate remediation and later accountability. RAIDT therefore matters because it gives the organisation a practical way to govern AI use where consequences are real, contested, and operationally specific.

• 01-Responsible_AI_for_Managerial_Decision-Making_Under_Uncertainty-V3
• 15-RAIDT-IS-Governance_M_v07
• 13-RAIDT-Evidence-Review_M_v10