• workshop_dense_100#slide 91

Assurance in RAIDT is the practice of making responsible-AI claims reviewable through a run-level evidence pack, a score profile, and documented controls tied to one configured use. RAIDT is explicit that governance should not rest on aspirations, supplier narratives, or periodic audit alone. Instead, it defines the run as the unit of governance and evaluates each important use through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). The score profile uses anchors 1=missing / 3=partial / 5=audit-ready so that evidence quality and governance readiness can be discussed in a disciplined, comparable form.

An example helps. Suppose a bank uses a generative assistant to draft customer-facing explanations in a sensitive service workflow. RAIDT assurance would require the exact run record: the prompt template, configuration, any retrieved policy content, the output, and the checks applied before the explanation was sent. The organisation would then review the score profile rather than relying only on the supplier's system-level documentation. If a complaint later arises, the bank has a bounded evidence object for challenge and review.

This matters in RAIDT because current governance instruments increasingly demand evidence, monitoring, and oversight, yet often leave the proof object under-specified. RAIDT supplies that missing operational layer. It makes assurance comparable across teams, workflows, and vendors; supports procurement, audit sampling, and complaint handling; and preserves the governance significance of influence methods as governance interventions. In short, RAIDT matters because it turns assurance from a general promise into a reviewable property of actual organisational use.

In finance, a customer disputes an explanation generated by an AI-enabled service workflow. Without RAIDT, the organisation might have only a policy document, a vendor assurance statement, and some partial logs. With RAIDT, reviewers can examine the run-level evidence pack for that exact interaction and see whether the workflow preserved the evidence needed for Responsibility, Auditability, Interpretability, Dependability, and Traceability.

That matters operationally. If the run shows incomplete provenance or weak review, the score profile exposes the weakness and gives the organisation a basis for remediation. If repeated runs reveal the same weakness, the institution can tighten controls, redesign the workflow, or revisit the supplier relationship. Assurance therefore supports not only accountability after the event, but continuous improvement in live organisational governance.

• 10-RAIDT_Policy_Pathways_M_V50
• 14-RAIDT-Policy-Motivation_M_v11
• 16-RAIDT-Audit-Accountability_M_v05