• workshop_dense_100#slide 57

In RAIDT, Responsibility means that a run was appropriate, bounded, safe, authorised and overseen in its actual context of use. It is one of the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), and it matters because generative-AI risk materialises in concrete runs rather than in abstract model descriptions alone. The foundations paper defines Responsibility as the extent to which a run satisfies ethical, legal and organisational obligations, while the evidence-review paper shows why such obligations must be grounded in reviewable evidence objects. Put simply, RAIDT asks not only, ?What did the system output?? but also, ?Should this run have happened in this way, under these controls, and with this degree of human oversight??

This matters in RAIDT because Responsibility links output use to accountability. A model card or policy statement may describe intentions, but it cannot by itself reconstruct a contested use event. A run-level evidence pack can. It records context, configuration, outputs, checks and oversight decisions so that a run can be challenged, justified or improved. Responsibility is therefore central to the RAIDT score profile: it prevents organisations from treating persuasive text as sufficient governance. It also makes room for influence methods as governance interventions, while insisting that such interventions be documented and monitored rather than trusted implicitly. In that sense, Responsibility is the pillar that turns responsible-AI aspiration into inspectable organisational practice.

A concrete example is the healthcare note-summarisation scenario. Suppose a GenAI assistant drafts a summary for a high-risk patient encounter. Responsibility matters because the draft may shape clinical understanding, documentation and follow-on action. Under RAIDT, a responsible run would include conservative instructions, a boundary that the output is draft support only, uncertainty and escalation guidance, and a recorded clinician decision about whether to use or amend the draft.

That matters operationally because, if the note is later questioned, the organisation can reconstruct the governed use rather than relying on vague assurances that staff were told to be careful. The evidence pack shows whether the system was used within an authorised and overseen process, which is precisely why Responsibility is indispensable in RAIDT.

• 08-RAIDT_Foundations_M_V50
• 13-RAIDT-Evidence-Review_M_v10