Australia’s approach to AI sovereignty should focus on operational capability rather than owning all AI technologies. The role of the new Chief AI Officers in the Australian Public Service can drive practical government AI adoption through shared missions, reusable capabilities and integrated safeguards, ensuring the country remains capable of selecting, adapting, governing and deploying AI to serve national priorities effectively and sustainably.

July 2026 marked a quiet but important deadline for the Australian Government. Every Commonwealth agency was required to appoint a senior Chief AI Officer to drive adoption and organisational change. At the same time, the Prime Minister created an Office of AI in his department and called for greater sovereign capability.

These initiatives create a useful test: will Australia treat sovereignty as owning more AI technology, or as possessing the capability to use global and domestic technology on Australian terms?

Sovereign AI does not equate to building a home-grown frontier model

Australia’s comparative advantage has rarely come from scale alone. In mining, medicine and financial services, success has emerged from combining imported and domestic technologies with specialised expertise, credible institutions, strong research networks and demanding users. AI will be similar. Australia does not need to produce every component of the AI stack to benefit from it, but it does need the capability to select, adapt, govern and deploy these technologies in ways that serve national priorities.

The Productivity Commission estimates that AI could add about four per cent to labour productivity over the next decade and notes that Australia begins with significant strengths: a highly educated population, skilled migration and world-class teaching and research institutions. The missing link is the institutional machinery that converts those assets into changed workflows, better services and cumulative learning.

The stronger goal is therefore operational sovereignty: the ability to select, adapt, govern and switch between AI systems without surrendering essential public functions to any single vendor or foreign state. Retaining targeted sovereign capacity in critical data, compute, cybersecurity, safety evaluation and mission-specific systems is essential to be competitive, but this does not equate to sovereign frontier-model development.

That distinction between technological ownership and capability matters because AI is a general-purpose technology. Its largest benefits will come not from possessing a model, but from the work to redesign it. The policy choice is therefore between passive dependence and capable interdependence. Passive dependence means adopting technologies on terms largely determined by overseas providers, without sufficient expertise to evaluate their performance, understand their risks or switch providers when circumstances change. Capable interdependence means remaining open to the world’s best technologies while retaining the knowledge, infrastructure and institutional capacity needed to use them on Australia’s terms.

For government, this requires more than procuring access to advanced models. It means protecting sensitive data, preserving competition among vendors, testing systems against Australian conditions and standards, and ensuring that agencies are not locked into a single technological ecosystem. We should focus on Australian applications in areas where local knowledge matters, from healthcare and social services to regulation, environmental management and emergency response.

The role of AI Chiefs in the Australian Public Service

The creation of the new Chief AI Officers recognised that Responsible AI can be used to make productivity gains across the public service. To ensure that this goal is realised, they must be harnessed for capability uplifts - not to add another layer of compliance or to become a loose network to share best practice.

The Department of Finance’s AI Delivery and Enablement function (AIDE), working with the Digital Transformation Agency and the Office of AI, could consider a common twelve-month mission to leverage these new roles. The work could begin with three service-wide missions where benefits can be measured and human judgement retained: reducing administrative processing times, improving access to government information and helping frontline staff navigate complex rules.

Agencies could nominate use cases against these missions, rather than launching disconnected pilots.  By publishing common baselines, the redesigned workflows, and the nominated accountable owner, the public benefit could be better captured and communicated.

Four practical steps are needed to move to implementation.

1. Map capability gaps. Chief AI Officer could assess the agency’s readiness across data quality, workforce skills, procurement, cybersecurity, evaluation and change management. A resulting whole-of-government capability map could help reveal where shared investment is more efficient than agency-by-agency spending.
2. Create protected experimentation capacity. A small, time-limited implementation fund would allow temporary multidisciplinary teams of service staff, data specialists, procurement officers, lawyers and workforce representatives to demonstrate user value and risk controls, not merely technical completion.
3. Reuse what works. A government repository of approved contract clauses could be maintained, including impact assessments, evaluation methods, reference architectures and deidentified lessons from failed and successful projects. This would reduce duplicated legal and procurement work while preserving agency accountability.
4. Measure outcomes, not activity. Each project could report a small set of comparable measures: time saved, service quality, staff uptake, error rates, user trust, vendor portability and the proportion of benefits reinvested in capability. Scorecard could shed light on the approaches that best demonstrate public value.

Australia already has much of the governance scaffolding needed to make this a reality. The updated Policy for the responsible use of AI in government requires agencies to register in-scope uses, assign accountable owners and complete impact assessments. The new AI Review Committee can advise on high-risk, sensitive and novel cases and the DTA’s has recently released new guidance for moving AI projects from proof-of-concept to scale.

These safeguards should be integrated at the design stage so that agencies can move faster with confidence, rather than treating governance as a final approval hurdle. The next step is to make scaling a collective APS capability. A procurement lesson in one department, a data standard in another and an evaluation method in a third should become reusable public assets, not remain local knowledge.

Making AI capability a whole-of-government mission

The Chief AI Officer network gives Australia a practical mechanism for developing this capability now. Properly organised, it could identify shared missions across government, coordinate common technical and procurement standards, and turn successful experiments into reusable public assets. Departments should not repeatedly solve the same problems in isolation. Evaluation methods, data-governance frameworks, procurement clauses, safety protocols and proven applications developed in one agency should be available to others.

This would also allow the APS to become a demanding lead user of AI. By setting high standards for transparency, security, interoperability and measurable public benefit, government can shape the behaviour of technology providers and stimulate the development of stronger domestic capabilities. Australian firms, universities and researchers would gain clearer pathways for working with government on nationally important problems, while lessons from public-sector deployment could inform responsible adoption across the wider economy.

This is a more durable form of AI sovereignty than owning a national model for its own sake. Sovereignty does not require control over every technology or freedom from all external dependence. It requires the capacity to make informed choices, manage unavoidable dependencies and ensure that technology continues to serve Australia’s public purposes. The real test is not whether Australia can build everything itself, but whether it can confidently decide what to adopt, what to protect, what to develop locally and when to change direction.

Vikas Kumar is Professor of International Business at the University of Sydney Business School and Editor-in-Chief of the Journal of International Management. His research examines global strategy, emerging markets and how firms and countries build capabilities in a changing global economy.

Image credit: iStock photo ID:2250412913