• qa_deck_100#slide 90 · Standards and policy interoperability
• integrated_82#Q4.15

RAIDT connects to the EU AI Act by acting as an evidence translation layer between statutory duties and inspectable organisational practice. In the policy-pathways paper, the Act is treated as a risk-based legal architecture whose key demands for high-risk AI include risk management, technical documentation, record-keeping and logging, transparency, human oversight, and expectations around accuracy, robustness, and cybersecurity. RAIDT does not replace those duties and does not claim legal uniformity. Instead, it maps them into a common evidence grammar so that the same governed use can be reviewed by regulators, buyers, auditors, and internal assurance teams without starting from scratch each time.

The practical link is that RAIDT treats the run as the unit of governance and turns one configured use into two inspectable artefacts: a run-level evidence pack and a score profile. The five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) operationalise different parts of the Act. Responsibility aligns most directly with Articles 9 and 14 on risk management and human oversight; Auditability and Traceability support Articles 11 and 12 on technical documentation and record-keeping; Interpretability supports Article 13 and the transparency logic associated with Article 50; and Dependability supports Article 15 on accuracy, robustness, and cybersecurity. In this sense, RAIDT helps organisations move from broad compliance claims to reviewable proof of how a specific use was governed in context.

A hospital deploying a clinical note-summarisation assistant can use RAIDT to show how one chest-pain summary was governed under AI-Act-style expectations. The run-level evidence pack records the prompt version, model configuration, retrieved context, output hash, clinician role, escalation rule, and reviewer sign-off. The corresponding score profile then shows whether the run had clear Responsibility, sufficient Auditability, intelligible explanations for different audiences, acceptable Dependability, and usable Traceability.

That matters because the EU AI Act asks for more than a generic policy file. If a regulator, auditor, or procurement reviewer asks how the system handled one high-stakes use, the organisation can present a concrete run record rather than a narrative assurance statement. RAIDT therefore connects the Act to practice by making one clinical use reconstructable, reviewable, and contestable.

• 10-RAIDT_Policy_Pathways_M_V50
• 14-RAIDT-Policy-Motivation_M_v11