S9.05 - Interoperability

S9.05 ? Interoperability

flowchart LR
    A[Fragmented governance demands
procurement, audit, assurance, policy] --> B[RAIDT
run-level evidence framework]
    H[Practical fields
healthcare, finance, public services, education] --> C[[Interoperability
one run evidence base reused across contexts]]
    B --> C
    C --> D[Evidence pack]
    C --> E[RAIDT score profile]
    C --> F[Reviewer reconstruction]
    C --> G[Governance readiness
reviewability, contestability, audit readiness]
    D --> G
    E --> G
    F --> G

? Star S9 - Policy, Standards and Assurance

Star context: Positions RAIDT within policy instruments, standards, assurance, procurement, audit, and organisational accountability by showing how one run-level evidence base can be interpreted across multiple governance settings without forcing those settings to become identical.

Academic picture

Definition / background

Interoperability, in the RAIDT context, means that the same run evidence can support multiple governance frameworks, review processes, and assurance audiences without requiring those frameworks to be collapsed into a single standard. It is therefore not merely technical compatibility between systems, nor a loose claim that different policies are "similar enough". It is an evidential and governance property: run-level records are structured in a way that lets different actors interpret the same underlying evidence for their own legitimate purposes.

Conceptually, interoperability matters because organisations rarely face only one governance regime. A generative AI deployment may simultaneously be subject to internal policy, procurement controls, audit requirements, sector guidance, management-system standards, and regulatory expectations. Without interoperability, each audience asks for separate documentation, often restating the same facts in different formats. This creates duplication, inconsistency, and a drift back towards assertion-heavy governance, where teams write documents for audiences rather than maintain evidence about actual runs.

Within RAIDT, interoperability belongs naturally because RAIDT treats the run as the unit of governance. A run records a specific use of a configured GenAI system for a specific task, time, and context. If that run is documented properly, the resulting evidence pack can be mapped outward to different frameworks, while the five-pillar score profile provides a concise view of governance performance. Interoperability therefore connects run-level evidence to broader policy and assurance ecosystems.

This also distinguishes interoperability from harmonisation. Harmonisation seeks alignment between standards or legal regimes themselves. Interoperability is less ambitious but more practical: it allows one evidence base to be used across distinct governance settings even when those settings retain different vocabularies, thresholds, and institutional logics. RAIDT benefits from this distinction because it focuses on operational evidence rather than pretending that all frameworks ask identical questions.

Why this concept matters

Interoperability solves a practical governance problem: organisations using GenAI are often asked to answer similar questions repeatedly for different audiences. When each audience demands a bespoke narrative, governance becomes slower, more expensive, and less reliable. Teams spend time rewriting compliance stories instead of improving the quality of evidence captured at the point of use.

It also avoids a common confusion in AI governance. High-level principles often look transferable, but review processes are not automatically interoperable. A policy statement that says a system is "safe" or "fair" does not help much if audit, procurement, assurance, and regulatory teams each need different forms of evidence. RAIDT addresses this by making the run record sufficiently structured that different reviewers can examine the same event from different angles.

If interoperability is missing, several risks appear. Evidence becomes duplicated across teams, contradictory versions of the same run can circulate, reviewers cannot easily reconstruct what happened, and assurance becomes dependent on presentation quality rather than evidential quality. In that setting, organisations may appear compliant while still being poorly prepared for challenge, audit, or incident review.

For organisations using generative AI, interoperability matters because governance has to operate across operational, managerial, and external settings. RAIDT uses run-level evidence to move from principle statements to operational governance. Interoperability is the mechanism that lets one well-structured evidence base support that move across multiple institutional contexts.

Key idea: Interoperability matters because RAIDT makes one run-level evidence base reusable across different governance audiences, reducing duplication while increasing consistency, reviewability, and audit readiness.

What this item enables

Reuse of the same run-level evidence across audit, procurement, standards review, and policy scrutiny.
Consistent interpretation of a GenAI run without rewriting its history for each governance audience.
Mapping between RAIDT evidence packs and external frameworks such as internal controls, assurance reviews, and standards-based assessments.
More defensible governance conversations because claims can be traced back to specific runs, settings, outputs, and human decisions.
Lower documentation burden for organisations that would otherwise maintain multiple overlapping evidence artefacts.
Better organisational learning because evidence gathered for one review can inform later reviews, incidents, procurement cycles, or policy updates.

Practical example / likely audience question

Audience question

If procurement, internal audit, and a standards review all ask different questions, why not just prepare separate documents for each instead of designing for interoperability?

Answer

The concern behind the question is understandable: separate documents seem easier because each audience uses its own vocabulary. However, that approach treats governance as a writing exercise rather than an evidence exercise. Separate documents usually restate the same underlying facts about the model, prompts, safeguards, outputs, human oversight, and exceptions, but they do so in slightly different ways. Over time, those versions drift apart.

The direct answer is that interoperability improves reliability as much as efficiency. If one run-level evidence pack is the shared source, then procurement can examine supplier and control implications, audit can test reconstructability and exception handling, and a standards review can assess whether the organisation followed a defined management process. The audiences differ, but the factual substrate remains the same.

A practical example is a GenAI assistant used to draft board briefings. Procurement may want to know which model is being relied upon and under what contractual controls. Internal audit may want to know whether prompts, outputs, approvals, and overrides were logged. An ISO-style review may want to know whether the organisation followed documented procedures for risk treatment and oversight. RAIDT handles this better than a generic AI governance approach because it starts from a run-level record rather than from generic policy claims. The evidence is captured once, then interpreted many times.

Practical example in RAIDT terms

Consider a public-service organisation using a generative AI assistant to draft responses to citizen enquiries. A single run involves a specific staff member, a configured model, a prompt template, an operational context, and an output that is reviewed before sending. The run-level governance issue is not just whether the tool exists, but whether this particular use can be reconstructed and justified later.

The evidence needed would include the task context, prompt or prompt template, model and version, data-sensitivity classification, applicable policy constraints, human reviewer identity, edits made before release, and any flags raised during quality or risk review. RAIDT would also record pillar-relevant signals such as whether the process was documented, whether the output could be explained to a reviewer, whether the run behaved consistently, and whether the trace from input to decision remained intact.

The affected RAIDT pillars are all five, but Auditability and Traceability are especially prominent because different oversight audiences must be able to inspect the same run. Responsibility matters because staff roles and approvals must be clear; Interpretability matters because reviewers need to understand why the output was accepted; Dependability matters because repeated runs should behave within known operational tolerances.

Interoperability improves governance readiness here because the same evidence pack can support a service manager, an internal auditor, a procurement renewal discussion, or an external policy review. Instead of recreating the story of the run each time, the organisation presents one structured evidential account that different audiences can read for their own purposes.

Detailed link to RAIDT

Interoperability links to RAIDT in four ways.

First, it supports RAIDT's core idea that governance should be grounded in evidence about actual uses of generative AI, not only in high-level principles or supplier claims.
Second, it depends on the run as the unit of governance, because a run is the smallest practical point at which context, configuration, action, review, and outcome can be captured together.
Third, it makes the evidence pack and score profile useful beyond RAIDT itself, since both can inform audit, procurement, assurance, and policy-facing review.
Fourth, it strengthens reviewability, contestability, audit readiness, and organisational learning by ensuring that multiple stakeholders can interrogate the same evidential base rather than parallel narratives.

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

In that chain, interoperability is the outward-facing property that allows RAIDT evidence to travel across institutional settings while still remaining anchored in the specific run from which the evidence originated.

Link to the five RAIDT pillars

Responsibility

Interoperability supports Responsibility because shared evidence clarifies who configured, approved, reviewed, and relied upon a given run. It reduces the temptation to blur accountability when documents are rewritten for different audiences.