What Canberra said this week
On 12 November, InnovationAus reported that Finance Minister Katy Gallagher said Australian-built AI models could eventually power GovAI, the Commonwealth's internal AI platform. Those models would replace the large US models the public service relies on today. According to the same report, Cabinet has discussed how a switch could be made, but the government's immediate focus is getting AI embedded in government services. Proactive Investors covered the same remarks and described the idea as a way to reduce the government's reliance on major US technology providers. Its report also said a sovereign GovAI could give Canberra more control over sensitive data, model behaviour and security.
Gallagher made the comments at the AI in Government Showcase. The Mandarin's report from the event says more than 110,000 Commonwealth public servants have completed basic AI training since last year's showcase, and more than 12,500 staff from 36 agencies are now connected to the GovAI Chat beta. The government hasn't published a timeline, a budget or a shortlist of Australian models for GovAI. The coverage we've seen doesn't name the US models GovAI uses now either.
Why a minister's remark carries weight
GovAI was built so agencies could change models. When the Whole of Government AI Plan was released in November 2025, ARN reported that GovAI would be vendor-agnostic, so agencies could pick models to suit the job and avoid lock-in. Gallagher's comments suggest the government now wants to use that flexibility for sovereignty reasons as well as for cost or performance. Because the application layer sits apart from the model, Canberra can talk about moving to Australian models without asking 12,500 beta users to learn a new tool.
Parliament is pushing in the same direction. The Joint Select Committee on AI was set up on 20 August 2026, and Clayton Utz notes that data sovereignty and national security feature prominently in its terms of reference. The committee's final report is due on 30 November 2026. Proactive Investors also reports that the government plans mandatory AI safety standards and a dedicated regulator. Taken together, these moves suggest Australian public sector buyers will soon be asking suppliers which model handles each request and who controls that model, on top of the usual questions about where data is stored.
What the GovAI move means for NZ businesses
NZ firms that sell to Australian agencies, or supply companies that do, should prepare for those questions now. A vendor that can only say "the data is hosted in Sydney" may struggle once buyers start asking where prompts are processed and whose model processes them.
In New Zealand, the local options for frontier models are still limited. Amazon Bedrock became available in the Auckland Region in March 2026, and AWS's own guide explains that requests using the Australian geographic inference profile can be served in Auckland or routed to Sydney or Melbourne. Under the global profile, requests can go to any supported commercial AWS Region. AWS says data at rest stays in the source Region. A NZ firm calling Claude through Bedrock in Auckland can therefore keep processing within Australia and New Zealand, but it can't guarantee processing in New Zealand alone. Australian processing is fine for plenty of workloads. It's a harder sell for health records, for Māori data held under iwi agreements, and for client contracts that specify New Zealand-only handling.
We think the larger lesson for NZ businesses is in GovAI's design. Most NZ organisations adopting AI have wired their workflows straight into one vendor's chat product or API, so prompts, integrations and staff habits all assume a single model. Suppose a board, client or regulator later asks for in-country processing. A business with that tight coupling then faces a rebuild where it could have made a configuration change. Firms that run models on their own hardware face the same risk from a different angle. Open-weight models are replaced every few months, and an on-premises setup locked to one model loses value quickly. Our private AI server work assumes the model will change during the hardware's life, and we'd apply the same assumption to any cloud deployment.
Two more facts are worth keeping in view. The GovAI change is still a possibility, with no decision announced, so NZ suppliers shouldn't redesign products around a policy that doesn't exist yet. And Australian sovereignty requirements won't automatically suit NZ needs. A model hosted in Canberra is offshore for a New Zealand organisation. A NZ firm that simply follows Australia's lead could end up with a setup that meets Australian expectations and still fails a New Zealand client's data residency clause.
Practical next steps
- List every AI service you use. For each tool or API, record which model it runs, where prompts and outputs are processed, and where logs are kept. Vendor marketing usually covers storage location and leaves processing location vague, so check the documentation.
- Read the processing clauses in your contracts. Look for whether the provider can route requests to other regions to manage capacity. Bedrock's global profile is one example of routing that's allowed by design.
- Sort your data by where it may be processed. A simple scheme works: data that can be processed anywhere, data that must stay in Australia or New Zealand, and data that must stay on infrastructure you control. Then link each class to the models approved for it.
- Put a gateway between your applications and the models. A routing layer lets you send sensitive requests to a local or regional model and everything else to a larger offshore model, and it makes changing models a configuration task.
- Test an open-weight model on hardware you control. Run it against one real workload so you know the quality and cost gap before a client asks you to close it.
- Watch the Australian committee report on 30 November if you sell into Australian government, and note any minimum standards it recommends for data sovereignty.
If you're setting rules for which data can go to which model, our AI consultancy team can help work out a classification and routing policy that fits your contracts and your clients' expectations.
Sources
- Cabinet discusses sovereign options for future GovAI models (InnovationAus)
- Tech Bytes: Canberra weighs sovereign AI models as Australia slips on AI competitiveness (Proactive Investors)
- GovAI driving public service productivity, says Gallagher (The Mandarin)
- Australian government's AI plan for public servants ensures sovereignty (ARN)
- From patchwork to playbook: the Joint Select Committee on AI and what it means for business (Clayton Utz)
- Run Generative AI inference with Amazon Bedrock in Asia Pacific (New Zealand) (AWS Machine Learning Blog)
- DataForge Insights feed
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help