What’s sovereign AI?

Sovereign AI is an AI service where your data, the processing of your questions and the AI model itself all stay inside Australia, under Australian control. For a business it comes down to one question: when someone in your company asks the AI something, does any part of that leave the country?

The term is often used by governments, to describe countries building AI capability they don’t have to rent from overseas. In business it has narrowed to something more practical. Can you say with confidence where your information goes when your people use AI?

Why “hosted in Australia” isn’t the whole answer

Plenty of AI products advertise Australian hosting. Often that’s true of the website, the application and the database, and untrue of the part that matters most. The question itself is passed to an AI model run by a large overseas provider, answered there, and sent back. The documents stayed in Sydney. The question, and everything typed into it, went overseas.

The five things that make it sovereign

• Storage: your documents, conversation history and settings are kept in Australia
• Processing: your questions are answered by a model running in Australia
• The model: your provider runs it in their own environment, so no overseas AI service is called
• Backups: copies are held in Australia too
• Training: your information never goes into training ChatGPT or any other public AI

Miss any one and the claim gets weaker. Backups held offshore, or a single step that calls an overseas service, is enough to change what you can tell a customer.

The Australian Signals Directorate makes the same point in its guidance on engaging with AI. AI systems “are often hosted in the cloud and may send data between different regions”, so organisations should check that any AI system they use can meet their data residency or sovereignty obligations.

Who actually needs it

Plenty of businesses don’t need all five. If AI is only being used to tidy up marketing copy, there’s little at stake. It matters most where:

• Customer contracts say where information may be processed
• You hold personal information about employees or customers
• Pricing, cost and tender information would hurt you if it got out
• You supply government, defence, health or critical infrastructure
• Your board or your insurer has started asking where AI is sending data

Australian Government agencies already apply this kind of test to hosting. The Hosting Certification Framework exists to help them find hosting services that meet “enhanced privacy, sovereignty and security requirements”.

There’s also the Privacy Act. Under Australian Privacy Principle 8, before a business discloses personal information to an overseas recipient, it must take reasonable steps to make sure that recipient doesn’t breach the Australian Privacy Principles, and in most cases the business stays accountable for what the recipient does with it. An AI tool that answers questions offshore can create that situation without anybody noticing. It’s worth getting your own advice on where you stand.

The trade-off, stated honestly

The big public AI services run very large models and improve quickly. A sovereign setup runs models hosted in Australia, which may be smaller. For most business work, like reading contracts, comparing figures, drafting and summarising, the difference is small, and having your business context matters far more than the size of the model. For some very general tasks the largest public models still have the edge.

How to check a provider’s claim

• Ask for the full path of a question, from the moment it’s typed to the moment it’s answered
• Ask whether any outside AI company’s service appears anywhere in that path
• Ask where backups and logs are kept
• Ask what happens to your data at the end of the contract
• Ask for the answers in writing

A provider that is genuinely sovereign will answer all five quickly, because the answers are simple.