C0.10 - Core_claim

C0.10 ? Core claim

flowchart LR
    A[High-level governance only:
principles, policies, model cards,
system cards, periodic audits] --> B[RAIDT:
run-level evidence framework]
    B --> C[[Core claim:
responsible GenAI governance
must be operationalised at the run level]]
    C --> D[Run-level evidence]
    D --> E[Evidence pack]
    D --> F[RAIDT score profile]
    E --> G[Reviewer reconstruction
and contestability]
    F --> H[Governance readiness
and organisational learning]
    I[Healthcare, finance, education,
public services, enterprise productivity] --> C

? Star C0 - RAIDT Core, Definition, Values, Claims and Innovation

Star context: Defines the project identity: RAIDT argues that responsible governance of GenAI in organisational work becomes credible only when grounded in run-level evidence, not only in principles, policies, model descriptions, or periodic audits.

Definition / background

The core claim of RAIDT is that responsible governance of generative AI in organisational work must be evidenced at the level where risk, judgement, and consequence actually materialise: the run. A run is one configured use of a GenAI system for a specific task, at a specific time, in a specific context. This claim does not deny the importance of model governance, organisational policy, or lifecycle assurance. Rather, it argues that these higher-level controls are insufficient on their own because they do not show what happened in a particular episode of use.

Conceptually, the claim responds to a persistent weakness in AI governance discourse. Many governance instruments describe what ought to happen, what a model is broadly capable of, or what an organisation intends to control. They are often less effective at showing what a human reviewer can inspect after a specific use event. RAIDT therefore repositions governance around the unit that can be reconstructed, evaluated, compared, and improved: the run.

Inside RAIDT, the core claim is the framework's central proposition. It justifies run-level evidence as the unit of proof, the evidence pack as the unit of review, and the score profile as the unit of structured judgement across Responsibility, Auditability, Interpretability, Dependability, and Traceability. Without this claim, RAIDT would remain a broad governance aspiration; with it, RAIDT becomes a method for producing practical evidence of governance readiness.

The item belongs in Star C0 because it states the thesis that holds the rest of the star together. The definition of RAIDT, the meaning of the run, the role of evidence over assertion, the emphasis on reviewability and contestability, and the account of innovation all depend on this claim being clear. It is therefore both a conceptual summary and a design rule for the whole project.

Why this concept matters

The core claim matters because it answers a simple but foundational question: where should responsible governance of GenAI be demonstrated? RAIDT's answer is that governance must be demonstrated where the system is actually used, configured, reviewed, and relied upon. If this is left vague, organisations can appear well governed at the policy level while remaining weak at the level of concrete operational use.

The concept also prevents a common confusion between governance declaration and governance evidence. A principle can declare that a system should be fair, safe, or accountable. A model card can describe intended use and known limitations. A periodic audit can review selected controls. None of these, by themselves, provide a reviewer with enough detail to reconstruct one situated use episode and assess whether governance expectations were actually met. The core claim therefore guards against over-claiming.

For organisations using GenAI, this matters in practical terms. Real harms and failures arise through specific uses: a misleading summary, an inappropriate recommendation, an untraceable prompt chain, an undocumented override, or a missing human review step. By focusing attention on runs, RAIDT provides a pathway from general governance ambition to operational governance practice.

Key idea: The core claim matters because it locates responsible GenAI governance at the point where use can be evidenced, reviewed, challenged, and improved.

What this item explains

Why RAIDT treats the run, rather than only the model or the organisation, as the decisive unit of governance analysis.
Why responsible AI governance requires concrete, inspectable evidence of specific use episodes rather than reliance on general statements alone.
Why evidence packs and score profiles are necessary outputs if governance claims are to be reviewable and comparable.
Why reviewability, contestability, and audit readiness depend on reconstructing what happened in a defined run.
Why organisational learning improves when repeated runs can be documented, compared, and scored over time.

Practical example / likely audience question

Audience question

Why is RAIDT's core claim stronger than simply saying that organisations should use AI responsibly?

Answer

The concern behind this question is that many governance frameworks already endorse responsible AI, so RAIDT may appear to be restating a familiar principle. The direct answer is that RAIDT does not stop at endorsement. Its core claim specifies where responsible governance must be made visible and how it can be evaluated. It argues that responsible use is not adequately demonstrated by policy commitments, model-level documentation, or occasional audit alone; it must be supported by evidence from the run in which the system was actually used.

A practical example makes the distinction clearer. An organisation may publish a strong policy on human oversight for GenAI-assisted report drafting. That policy is useful, but it does not show whether one particular report was generated with the approved prompt template, whether the correct model version was used, whether a human checked the output, whether sensitive data were handled properly, or whether downstream edits were logged. RAIDT's core claim insists that this level of detail matters because it is the level at which governance succeeds or fails in practice.

RAIDT handles this issue better than a generic AI governance approach because it links the claim directly to operational artefacts: run-level evidence, evidence packs, and the five-pillar score profile. The result is not merely a statement that governance should exist, but a framework for showing that governance was enacted in a particular use episode.

Practical example in RAIDT terms

Consider a hospital using a GenAI system to draft discharge summaries for clinicians. The use case appears low-friction, but the governance question is not whether the hospital has an AI policy in general. The run-level question is whether one specific discharge-summary draft, produced for one patient encounter by one clinician at one time, was generated under the correct safeguards.

In RAIDT terms, the run-level issue includes the prompt template used, the model and version invoked, the clinical context, whether patient data were minimised or masked appropriately, whether a clinician reviewed the output before release, and whether any material corrections were made. The evidence needed would include run metadata, prompt and output records where policy permits, reviewer notes, approval status, exception flags, and provenance links to the surrounding workflow.

The most affected RAIDT pillars would be Responsibility, Auditability, Dependability, and Traceability, with Interpretability supporting explanation of why the output was accepted or amended. The core claim improves governance readiness here because it shifts the hospital from saying "we govern GenAI responsibly" to showing how a concrete clinical use can be reconstructed, assessed, and improved.

Detailed link to RAIDT

Core claim links to RAIDT in four ways.

First, it states RAIDT's central thesis: responsible governance becomes credible only when it is demonstrated at the level of actual GenAI use.
Second, it anchors the framework in the run and makes run-level evidence the basic unit of governance proof.
Third, it explains why RAIDT produces evidence packs and score profiles rather than relying on unstructured assurance claims.
Fourth, it connects the framework to reviewability, contestability, audit readiness, and organisational learning by making specific uses reconstructable and comparable.

Core claim ? Run-level evidence ? Evidence pack ? RAIDT score profile ? Governance readiness

This chain is central to RAIDT because the claim is not an isolated statement. It is the logic that turns a theory of governance into a practical method for documenting, scoring, reviewing, and improving organisational GenAI use.

Link to the five RAIDT pillars

The core claim shapes all five RAIDT pillars, but it has its strongest direct force in Auditability and Traceability because both depend on being able to reconstruct a run.

Responsibility

The claim supports Responsibility by requiring that each run has a defined purpose, user role, decision context, and accountability structure. Responsibility is weakened when AI use is discussed only in generic organisational terms.