Five questions that test any sovereign AI claim
Many providers now sell their AI as sovereign. Put these five questions to yours, and if every answer is you, your AI is independent.
- Who holds the weights? The providerThe provider can change or withdraw the model you use. You doYou hold them under the base model’s licence, and they leave with you.
- Where does it run? In AustraliaYour data is stored in an Australian data centre. Melbourne, or your own data centreConnected or air-gapped, with your data and IP kept in your instance.
- Whose law reaches it? American law, wherever it’s storedThe provider can be required to hand over data in its “possession, custody, or control” (US CLOUD Act, 18 U.S.C. § 2713). Only the law you already answer toWe’re Australian-owned with no foreign parent, and the contract puts your data under Australian law.
- Who decides when it changes? The providerThe provider updates the model on its own schedule. You doA new version goes live only when you approve it.
- Who can switch it off? The provider, or a governmentOn 12 Jun 2026 a US order made a frontier lab suspend its newest models for foreign nationals, and it switched them off for every user. You doWeights you hold, on hardware you control, stop only when you or Australian law stop them.
Onshore, but not independentIndependent
Why it mattersDefinition
What sovereign AI should mean.
- Sovereign AI /ˈsɒv.rɪn eɪ aɪ/ noun
- AI that an organisation or a nation controls itself, because it holds the model’s weights, decides when the model changes, runs it on hardware within its reach and answers only to its own law, so that no one else can change it, read its data or switch it off.
Each model carries its lab’s defaults, and the Matilda training system retrains them on your law, markets and institutions.
The evidence · Why we’re getting right behind open weights
Where to start