• workshop_dense_100#slide 84

Within RAIDT, the EU AI Act is best understood as the statutory governance context that gives legal force to concerns which earlier responsible-AI work often expressed only as principles. The papers characterise the Act as a risk-based regime that makes documentation, record-keeping, transparency, human oversight, and risk management organisationally significant duties, especially for high-risk systems. That matters because RAIDT is designed for exactly the policy-to-operation gap exposed by the Act: organisations may know they must govern AI, yet still lack a standard proof object for showing how one configured use was actually controlled.

RAIDT matters because it supplies that proof object. By treating the run as the unit of governance, it converts the Act's broad obligations into a run-level evidence pack and a score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). This does not mean RAIDT determines legality by itself. The papers are explicit that scoring supports reviewability and comparability rather than substituting for legal judgement. Its value is operational: it helps organisations populate technical files, support conformity assessment, enable internal audit sampling, structure procurement questions, and preserve evidence for challenge and contestation. In short, the EU AI Act explains why evidence-bearing governance is now mandatory, while RAIDT explains how that requirement can be made inspectable, repeatable, and proportionate in day-to-day use.

A bank using generative AI to draft an adverse-action explanation after refusing credit illustrates why the EU AI Act matters in RAIDT. The letter must be understandable, contestable, and linked to the actual decision basis. A fluent but weakly evidenced explanation would be risky because it could appear transparent while hiding vague or invented reasons.

With RAIDT, the run-level evidence pack records the decision reason codes, the explanation template identifier, links to the policy fields supporting each reason, any uncertainty statement, and the human review step before release. The score profile then shows whether the run meets the expected standard for Responsibility, Interpretability, Auditability, and Traceability. In practice, this lets the organisation answer a regulator, auditor, or complainant with a reconstructable record of one governed use, which is exactly why the EU AI Act is so consequential for RAIDT.

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