Russell Buzby

On 30 June, days from now, every non-corporate Commonwealth entity is meant to have a Chief AI Officer in place. The Department of Finance set the requirement in 2025, and the instinct behind it is sound: put a senior, accountable leader in charge of AI adoption before the technology is everywhere rather than after. The catch sits in the fine print. The Chief AI Officer is to be drawn from each agency’s existing senior leadership, no new funding is attached, and most agencies that have moved are simply adding the title to an executive who already has a full job.

That design choice tells you most of what the mandate can and cannot do. Appointing from existing ranks with no new money is the cheapest available form of compliance, and it produces a designation rather than a capability. A chief information officer or chief data officer adds Chief AI Officer to the business card, the box is ticked, and the workforce capability and data plumbing that AI adoption actually requires stay exactly where they were the day before. The title changes. The capacity does not.

The uptake numbers, such as they are, point the same way. When iTnews asked agencies late in 2025 how they planned to meet the requirement, only fourteen would say. Seven of those intended to fold the role into an existing senior executive’s remit. Three were trying to appoint a standalone Chief AI Officer, and four were still waiting on advice before deciding. InnovationAus described the broader public-sector AI uplift in late May as patchy, while noting that progress is being made. Patchy is the right word for a reform that asks individual agencies to find capacity they were given no new resources to build.

This is the implementation gap, the thread that runs through most of what I write about public sector technology. An organisation can hold a complete set of frameworks and still lack the workforce capability and underlying data arrangements to execute them, which is why the opportunity for anyone advising government sits in the doing rather than the designing. The designing is largely finished. The National AI Plan and a whole-of-government policy for responsible use arrived first, an APS AI Plan followed, and in April the government tabled its response to the Senate Select Committee on Adopting Artificial Intelligence. What remains scarce is the operational capacity to turn those documents into delivered services, and a re-designated executive with no new budget does not create it.

Tasmania shows how the same gap behaves under fiscal pressure. The state announced roughly $61 million in digital transformation spending, including a $4 million AI sandbox, in the same budget that imposed around $1.47 billion in public sector spending and staffing cuts. Using technology as a fiscal lever to offset workforce reductions is an attractive proposition for any treasury. It also rests on assumptions that are rarely tested rigorously before the savings are booked: that the technology is ready, and that the remaining workforce can absorb the change while service continuity holds through the transition. When those assumptions fail, the cuts take effect immediately while the promised productivity gains stay deferred and uncertain. Citizens absorb the difference.

Queensland is already meeting the friction that arrives once adoption is underway. The state’s privacy commissioner and AI advocates have been openly debating the tension between the productivity gains from tools like the government’s QChat assistant and the growing privacy risks as data sharing expands across agencies. This is the predictable next stage. Policy ambition meets regulatory obligation at the point of actual use, and the unresolved questions about data protection that looked abstract in a strategy document become concrete the moment a public servant pastes something sensitive into a chat window.

The deeper risk sits on the citizen’s side of the counter. A new Australian framework published in late May, CLEAR-AU, makes the point that most of our responsible AI frameworks are institution-facing and reactive. They tell agencies how to govern their own use. They do very little to give citizens proactive tools to understand or challenge an automated decision that affects them. If government AI adoption accelerates without a matching investment in citizen-facing accountability, trust erosion stops being a risk and becomes a forecast. I have argued separately, in writing about the pattern behind automated decision failures from Robodebt to overseas welfare-fraud systems, that the accountability scaffolding has to be built before deployment rather than retrofitted after a scandal. The CLEAR-AU diagnosis is the same problem viewed from the public’s vantage point.

It would be comforting to assume the tooling will mature fast enough to cover the capacity gap. The international evidence counsels patience. A study of AI in public administration in Karnataka found service delivery genuinely advancing while bias and democratic accountability concerns deepened under transparency frameworks that were never built for algorithmic systems. A separate evaluation of standard responsible-AI tools such as Model Cards and ALTAI found they lacked sufficient documentation and practical guidance to be consistently useful when applied to Portuguese-language models. The instruments we reach for to make AI accountable are themselves immature in exactly the settings where the stakes are highest. Leaning on them to compensate for thin human capability is optimistic.

The most uncomfortable point lies in which chairs get filled, and how. An empty chair would at least be honest about a lack of capacity, whereas a chair filled by re-designation does the opposite: it signals a readiness the agency does not have, because a senior leader now formally owns AI with no new means to act on it. That move is a familiar one in Australian AI policy. I have argued before that launching AI.gov.au gave the Commonwealth a useful piece of furniture rather than the room itself, and that the country’s broader AI problem is one of organisational learning rather than technology. A nominal Chief AI Officer is that same furniture at the level of a job title: the appointment is real, but the capability it implies is mostly not.

A mandate measures intent. Appointing from existing staff with no new funding measures something smaller: a willingness to be seen to comply by the deadline. That deadline, on 30 June, will mostly be met on paper. Whether anything actually changes on 1 July is the question that matters, and for most agencies the honest answer will be “not much”.

References

Department of Finance. (2025). Establishing Chief AI Officers for the APS.

Digital Transformation Agency. (2025). Australian Public Service AI Plan 2025; Policy for the responsible use of AI in government (v2.0).

iTnews. (2025, December). Federal chief AI officer roles set to go to existing APS staffers.

InnovationAus. (2026, May 28). Public sector AI uplift “patchy” but progress is made.

InnovationAus. (2026, May 24). Tasmania looks to tech uplift to ease budget pain.

The Mandarin. (2026, May 27). Queensland’s digital mandarins talk AI.

CLEAR-AU. (2026, May 25). A citizen-facing responsible AI framework for Australia [Preprint]. Zenodo/OpenAlex.

Australian Government. (2026, April 1). Response to the Senate Select Committee on Adopting Artificial Intelligence.

AI in public administration in Karnataka [Preprint]. (2026, May 31). OpenAlex.

Evaluation of AI ethics tools for Portuguese-language models [Preprint]. (2026, May 29). arXiv.