6Operating Model (accountability and flow)

If strategy is the spark, the operating model is the engine that turns it into coordinated movement. It defines who is accountable for AI, how decisions about AI are made and escalated, how the three lines of assurance work, and what capability the agency needs to govern AI well. The Policy for the responsible use of AI in government makes much of this mandatory through its requirement for accountable officials and a use-case register. This section sets out an operating model that satisfies the policy and is proportionate to a public-sector agency.

6.1Four layers of accountability

The operating model has four layers, each with a single, named point of accountability. The layers separate the setting of strategy and appetite from the governance of use cases, the running of use cases, and the operation of the secure platform beneath them. This separation is what prevents the confusion in which no one can say who owns a decision.

6.2Accountable officials and decision rights

The policy requires designated accountable officials, who answer for implementing it and sit in the strategic layer, and, by 15 December 2026, an accountable use-case owner for each in-scope use case, sitting in the operational layer. Since July 2026 the APS AI Plan has also required every agency to have a Chief AI Officer; in most agencies the CAIO and the accountable official for AI strategy are naturally the same appointment, and this framework assumes so unless the agency deliberately separates the roles, in which case their respective decision rights should be written down on day one. This framework treats accountability as a chain that cannot be broken: the accountable authority is answerable for the agency’s AI overall; the AI Governance Committee is accountable for the integrity of the governance process; and a single named official is accountable for each use case, including its impact assessment, its controls and its outcomes. Responsibility for doing the work can be widely shared, but accountability, the duty to answer for it, cannot be delegated to a vendor, a committee, or a machine. Appendix C provides a RACI that assigns every key AI governance activity across these roles.

6.3The AI Governance Committee

The AI Governance Committee is the operating heart of the model. Chaired by the accountable official for AI strategy, with cross-functional membership spanning service delivery, data, security, privacy, legal, and the agency’s risk function, it approves use cases and their risk tiers, owns the use-case register, reviews impact assessments, approves transparency statements, and acts as the go/no-go authority at the assurance gates in Section 10. It is deliberately a decision-making body, not a discussion forum: every use case has a single accountable owner who brings it to the committee, and the committee’s role is to assure, challenge and decide. Appendix C includes indicative terms of reference.

6.4Three lines of assurance

Sound governance separates those who own a risk from those who oversee it and those who independently assure it. The framework adopts the established three-lines model, applied to AI.

6.5Capability and the new professions

AI governance asks for skills the public service has not historically organised around. Renly names six emerging capabilities that, while they need not all be full-time roles in a single agency, must exist somewhere in the operating model. We recommend mapping each to an accountable home, whether an individual, a function, or a panel partner, rather than assuming it will happen by default.

Alongside these specialist capabilities, the broader workforce needs baseline AI literacy, proportionate to role, so that staff can use AI tools well, recognise their limits, and know when to escalate. The framework treats capability uplift as a deliverable of the operating model, not an afterthought.

6.6Culture: from gatekeepers to participants

Finally, the operating model carries a cultural shift. In a world of AI at machine speed, humans cannot be in the path of every decision; if they try, they become the bottleneck rather than the governor. The more durable posture is to move humans from gatekeepers to participants: to design clear points at which human judgment is required, to equip people to exercise it well, and to free them from the points where it adds only delay. Section 8 builds this into the design of individual systems. At the level of the operating model, it means investing in the few human decisions that matter most, and automating the assurance of the rest.

6.7Operating model outputs