S2.04 - Accountability

S2.04 ? Accountability

flowchart LR
    A[Vague responsibility claims
Unclear ownership
Weak evidence] --> B[RAIDT
Run-level evidence framework]
    B --> C[[Accountability
Evidence-backed answerability]]
    H[Public services
Healthcare
Legal
Cybersecurity
Enterprise productivity] --> C
    C --> D[Run-level 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 S2 - Governance Meaning and Problem Context

Star context: Clarifies governance as oversight, control, accountability, reviewability and continuous improvement rather than a vague ethics label. In RAIDT, accountability turns governance from a general expectation into a demonstrable capacity to answer for a specific GenAI run.

Academic picture

Definition / background

Accountability means that identifiable actors can answer for how a run was configured, used and reviewed, and that they can do so with evidence rather than retrospective assertion. In organisational GenAI governance, this matters because the mere existence of policies, role titles, or approval chains does not by itself show that a particular use of a model was governed well. Accountability is demonstrated when the organisation can reconstruct and justify a specific instance of use.

Conceptually, accountability is closely related to responsibility, but the two are not identical. Responsibility assigns duties, ownership, or expected conduct. Accountability adds the requirement that those actors can be questioned and can justify action in an inspectable way. It is also distinct from auditability and traceability. Auditability concerns whether review is possible; traceability concerns whether events and artefacts can be followed through the process. Accountability depends on both, but it is the broader governance condition in which answerability is meaningful.

This is why accountability belongs inside RAIDT. RAIDT treats the run as the unit of governance, so accountability is attached to one configured use of a GenAI system for a specific task, at a specific time, in a specific context. The run-level evidence pack gives accountable actors the material needed to answer questions, while the five-pillar score profile indicates whether the run was governed in a way that supports credible accountability. In this sense, RAIDT does not treat accountability as an abstract ethical aspiration; it treats it as an evidence-backed organisational capability.

Why this concept matters

GenAI governance often fails at the moment when a senior manager, auditor, regulator, or affected stakeholder asks a simple question: "Who can explain what happened here?" Without a structured answer, organisations fall back on vague claims such as "the human was in the loop" or "the vendor supplied the model". Those claims diffuse responsibility, obscure decision pathways, and make post hoc learning difficult. Accountability matters because it gives governance a practical location, a reviewable record, and an organisational owner.

In RAIDT, accountability also solves a methodological problem. Many governance frameworks stay at the level of principles and controls, but do not show how an organisation evidences them for a particular use of AI. By grounding accountability at run level, RAIDT helps move from policy language to operational proof. That matters for PhD supervision, viva defence, implementation design, and practitioner uptake because it explains how governance can be examined rather than merely declared.

Key idea: Accountability matters because RAIDT makes answerability demonstrable for a specific GenAI run, rather than leaving it as a vague organisational claim.

What this item enables

It enables named actors to justify how a particular GenAI run was set up, reviewed, and acted upon.
It enables governance discussions to move from generic responsibility statements to inspectable run-level evidence.
It enables reviewers to connect organisational roles with prompts, model choices, inputs, outputs, checks, and downstream decisions.
It enables escalation when a run exceeds policy, competence, or risk thresholds.
It enables learning after failure, because the organisation can examine not only what went wrong but also who could have intervened and with what information.
It enables stronger audit readiness, contestability, and internal governance confidence across repeated uses of GenAI.

Practical example / likely audience question

Audience question

Who is accountable?

Answer

The organisation assigns roles; RAIDT provides the evidence by which those roles can answer questions.

The concern behind this question is usually that GenAI appears to blur ownership across vendors, internal users, managers, policy teams, and reviewers. The direct answer is that accountability for organisational use cannot be outsourced simply because a third-party model is involved. Vendors may carry design, documentation, or contractual obligations, but the deploying organisation remains accountable for how a specific run is initiated, configured, reviewed, and used in its own work context.

A practical example is a team using a large language model to draft a client-facing summary. If the output omits a critical qualification and the summary is sent onward, the organisation must be able to show who initiated the run, what instructions were given, whether the task was in scope, what review was performed, and who approved use of the output. RAIDT handles this better than generic AI governance because it does not stop at saying there should be human oversight. It captures the evidence that shows which human, under which conditions, with what review basis, and with what governance consequences.

Practical example in RAIDT terms

Consider a public-service setting in which a local authority caseworker uses a GenAI system to draft a housing-benefit appeal summary. The run-level issue is that the model compresses supporting evidence and understates a claimant's medical circumstances, making the draft unsuitable if used without careful review.

For accountability, the organisation needs evidence of the task purpose, the prompt, the model and version used, the source documents supplied, the user identity or role, the time of the run, the review outcome, any edits made, the approval or rejection decision, and the reason that decision was taken. It also needs to know whether the case was suitable for GenAI support under policy and whether escalation was required.

The most affected RAIDT pillars are Responsibility, Auditability, and Traceability, with important implications for Interpretability and Dependability as well. If this evidence is present, a supervisor can reconstruct what happened, determine whether policy was followed, and identify whether the failure arose from task selection, poor prompting, weak review, or over-reliance on the output. Accountability therefore improves governance readiness because it turns a potentially disputed event into a reviewable and learnable case.

Detailed link to RAIDT

Accountability links to RAIDT in four ways.

First, it connects directly to RAIDT's core idea that governance should attach to the run rather than remain at the level of broad principle. A run is where real organisational action occurs, so a run is where answerability must be established.

Second, it depends on run-level evidence. Without a record of configuration, context, interaction, review, and decision handling, accountability collapses into unsupported narrative.

Third, it is operationalised through RAIDT outputs. The evidence pack gives the documentary basis for answering questions about a run, while the score profile indicates whether the run met the conditions that support robust accountability.

Fourth, it strengthens wider governance capabilities such as reviewability, contestability, audit readiness, and organisational learning. Accountability is what allows these wider functions to be exercised in practice rather than merely stated in governance documents.

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

This chain matters because RAIDT turns accountability from a high-level governance ideal into a structured evidential pathway. If accountability is weak at run level, the evidence pack will be thin, the score profile will expose weaknesses, and governance readiness will be correspondingly limited.

Link to the five RAIDT pillars

Responsibility

Accountability is most visibly linked to Responsibility because it clarifies who is expected to act, review, approve, or escalate within a run. However, RAIDT adds that responsible actors must be able to justify those actions with evidence.