What Is Sovereign AI? Meaning, Examples and Five Tests for Your Organisation
Sovereign AI means you control where your AI runs, who holds the data and whether you can switch or unplug it. Five tests to check yours.

On this page
In short: sovereign AI means being able to build, run and govern AI on your own terms: on infrastructure you control, with data and models that stay under your jurisdiction, and with the freedom to change provider or switch the AI off. The term is also used for national AI capacity. For an organisation, it means control over its own AI agents, automations and applications.
What is sovereign AI?
Sovereign AI is control over the AI you use: where it runs, what data and models it uses, who approves what it does, and whether you can switch it off. The term has two meanings:
| National sovereign AI | Organisational sovereign AI | |
|---|---|---|
| Who it’s about | A country or region | A company, public body or charity |
| The goal | Build AI capacity at home: compute, data, skills, models | Control the AI you use: where it runs, what it sees, who decides |
| Typical investment | Data centres, national models, research funding | Self-hosted platforms, data residency, governance |
| The risk it answers | Dependence on another country’s technology | Dependence on one vendor’s pricing, policies, uptime and jurisdiction |
Many definitions you’ll find describe the national version. This guide is about the second meaning: what sovereign AI means for an organisation that has to answer to a regulator, a board, citizens or donors.
Why sovereign AI matters
Jurisdiction follows the provider, not just the data centre. The US CLOUD Act (2018) allows US authorities to require US-based providers to hand over data they control, even when it is stored outside the US. Hosting in an EU region does not, on its own, answer that question.
Data transfer rules have changed before, quickly. In the Schrems II judgment (Case C-311/18, 16 July 2020), the Court of Justice of the EU struck down the EU-US Privacy Shield. Organisations relying on it had to find another legal basis for their transfers.
Regulation needs evidence you can produce. GDPR, ISO/IEC 42001, the EU AI Act and any country’s own laws ask who decided, on what data and under which policy. Under the Digital Omnibus on AI, the EU AI Act’s high-risk rules apply from 2 December 2027, and from 2 August 2028 for AI built into regulated products; check the final text. Evidence is easier to produce when the system and its records sit on infrastructure you control.
Your processes now depend on someone else’s roadmap. When a core process runs through one AI provider, that provider’s pricing, policies and uptime become yours. If they change, your operations change with them.
Five tests for sovereign AI
Use these five questions on any AI system you run or plan to buy. These tests are our own framework, not a legal standard.
| # | Test | What sovereign looks like | Red flag |
|---|---|---|---|
| 1 | Where does it run? | On infrastructure you choose: your cloud, on-premises or in-country | “Our cloud only”, with no say in region or provider |
| 2 | Who holds the data and the models? | Your data, prompts and model outputs stay with you | Your data is used to train someone else’s model, or you can’t export it |
| 3 | Can you change the AI provider? | Any provider, swapped without rebuilding the system | The system only works with one vendor’s model |
| 4 | Who decides? | Named people approve releases and outcomes that matter | The AI acts and approves its own changes |
| 5 | Can you unplug it? | The system keeps working, or stops in a controlled way, with the AI switched off | Switching off the AI stops the service |
If a system fails test 1 or 2, you have a data sovereignty problem. If it fails 3 or 5, you have a dependency problem. If it fails 4, you have a governance problem. In this framework, sovereign AI needs all five.
An example: a council chatbot
This is a hypothetical illustration, not a customer story. A council runs a residents’ chatbot on one vendor’s hosted model, in a region the council didn’t choose. Test 1 fails: no say in where it runs. Test 3 fails: it only works with that vendor’s model. Test 5 fails: switch the model off and the service stops. Moving the chatbot to infrastructure the council controls, with the model behind a swappable connection and a fallback to a human-staffed form, should turn those three fails into passes.
Sovereign AI vs sovereign cloud vs open-source AI
These terms overlap, but none of them is the whole answer.
| What it gives you | What it doesn’t | |
|---|---|---|
| Sovereign cloud | Data hosted in a chosen jurisdiction, often by a local operator | Control over which AI models you use, or who approves what they do |
| Open-source AI | Models and code you can inspect, run and modify | Hosting, governance, approvals or evidence |
| Sovereign AI | Control over where AI runs, what data it uses, which provider it uses and who decides | It still needs people to run it well |
Common misconceptions about sovereign AI
“Hosting in an EU data centre makes it sovereign.” Not on its own. Who operates the service, and under which law, matters as much as where the servers sit.
“Sovereign AI means never using US models.” No. It means you choose the provider, you can switch, and your data stays under your control while you use it.
“Open source solves it.” Partly. Open source gives you code you can inspect and run. You still need governance: owners, approvals, guardrails and records.
How to move towards sovereign AI
- List your AI systems. Include agents, automations, applications and AI features switched on inside software you already pay for. Our guide to shadow AI covers the ones nobody approved.
- Run the five tests on each. Mark each system pass, partial or fail.
- Start with the riskiest fail. For example, a system that handles personal or special-category data on infrastructure you don’t control.
- Separate the model from the system. Build so the AI provider can be swapped. Where a tested workflow would do the job, use it instead of an agent: see AI agents vs workflows.
- Put people in charge. Name an owner and an approver for every system. Our AI governance framework guide sets out the roles.
- Write down your exit. For each system, record how you would change provider or switch the AI off, and test it once.
How Planet Crust @Human approaches sovereign AI
Planet Crust @Human is The Sovereign AI Enterprise System. You build compliance-led AI agents, automations and applications in one place, governed by design, on infrastructure you control.

Here is how @Human answers the five tests:
- Where does it run? Self-hosted, as software you run yourself: your cloud or on-premises. For the public sector, everything can be hosted in-country.
- Who holds the data and models? Your data and models stay with you.
- Can you change provider? @Human works with any AI provider you choose.
- Who decides? Accountable roles and signed approval gates. AI agents can’t sign a gate.
- Can you unplug it? Yes. @Human can run with AI switched off. Where both would work, it picks programmatic automation before agentic, which also means lower token costs.
Watch: self-hosted, sovereign AI with @Human (1:30)
This 90-second film shows these answers on screen: installing @Human on your own server, choosing your AI model or switching the AI off, and a named person signing the gate when an AI agent needs approval.
@Human is open source, and every @Human instance runs its own MCP server: an AI assistant you connect works always as you and always within your permissions. It is ready for the EU AI Act, ISO/IEC 42001 and GDPR, and for any country’s laws and regulations.
@Human gives you the controls and the evidence. How sovereign your AI estate becomes still depends on the choices you make with them. Pricing is pay what you can, from €1 per month, and not-for-profits get a free build.
Frequently asked questions
What does sovereign AI mean in simple terms?
It means you stay in control of your AI: where it runs, who holds the data, which provider it uses, who approves what it does, and whether you can switch it off.
What is the difference between sovereign AI and sovereign cloud?
Sovereign cloud is about where data is hosted and under which law. Sovereign AI goes further: it also covers which models you use, whether you can change provider and who decides what the AI does.
Is sovereign AI only for governments?
No. Any organisation that answers to somebody, from public bodies and healthcare to financial services, education and not-for-profits, can apply the same idea to its own AI systems.
Does sovereign AI mean building your own model?
No. Most organisations use existing models. Sovereignty comes from controlling where they run, what data they see and whether you can replace them.
How do I know if my AI is sovereign?
Run Planet Crust’s five tests (our own framework, not a legal standard): where it runs, who holds the data and models, whether you can change provider, who decides, and whether you can unplug it. In this framework, a system needs to pass all five.
Your next step
Pick the AI system your organisation depends on most. Answer the five tests for it today, in one line each. If you can’t answer test 3 or test 5, that’s the dependency to fix first. To see a sovereign, self-hosted AI platform working with your use case, Book a free walkthrough. A real person replies within one working day.
This article is general information, not legal advice.
Sources
- Regulation (EU) 2024/1689, the EU AI Act, EUR-Lex.
- Regulation (EU) 2026/1744, the Digital Omnibus on AI, EUR-Lex, 2026 (high-risk dates).
- Regulation (EU) 2016/679, the GDPR, EUR-Lex.
- Clarifying Lawful Overseas Use of Data (CLOUD) Act, US Department of Justice, 2018.
- Judgment in Case C-311/18, Data Protection Commissioner v Facebook Ireland and Maximillian Schrems (Schrems II), Court of Justice of the EU, 16 July 2020.
- @Human product facts: @Human for Enterprise, Why @Human, @Human for the public sector and @Human MCP, Planet Crust, November 2026.