Sovereign AIIn plain terms

What is sovereign AI? A plain answer for UK businesses.

Sovereign means holding supreme or final power. Sovereign AI is AI where the final say over the model, the data and the computing power rests with the people it serves: the country whose law applies, the company that owns the work, and the individual the information is about. For a business, that means your client files, and the work done on them, stay where you decided, under your control, and you can prove it.

Compare the two options Talk to us about sovereignty

✓ Country, company, individual✓ Your environment or ours✓ A receipt for every step

Updated 5 Nov 2026

A padlocked steel door beside a corridor of servers

Where it ranUnited Kingdom · operator named

ReceiptModel, rules and cost recorded

01 · What it means

What sovereign AI means.

There is no single agreed definition of sovereign AI yet. Governments tend to mean national capability: in the UK, the Sovereign AI Unit is a £500m government-backed fund that invests in British AI companies, and the UK Compute Roadmap describes sovereignty as the ability to act independently and effectively where it matters most. Suppliers often mean where a server sits. We start from the word itself. A sovereign holds supreme or final power, so the question for any AI is who has the final say over three things: the model (the weights that do the reading and writing), the data (what goes in and what comes out) and the compute (the machines that run it).

Our answer is that the final say belongs to three parties, each over its own part: the country whose law applies, the company that owns the work, and the person the information is about. Each has a degree of control, a degree of residency and a degree of ownership.

UK lawEU lawWherever you trade

Your frameworksYour playbooksYour data

Sam Patela model seesPerson A

ControlUsed for payroll, with consent

ResidencyKept in the UK

OwnershipCan see, correct or erase it

CountryWhose law applies

ControlYour business is governed by the law of the country that owns it, and by every law that reaches its work. Those rules are applied before anything moves.

ResidencyStored and processed in the UK and EU by default. Personal and confidential work stays in its home jurisdiction.

OwnershipWho owns the infrastructure matters as much as where it is, because an operator owned abroad can be compelled by its home country's law. We name the operator of every place your data can go.

Law is a filter, never a score.

Where your work runs is decided by rules, never by a model. The law removes places first; only then do cost and speed choose among what's left. If nowhere is allowed, the work is refused, and the refusal is recorded.

02 · Why now

Why UK professional services businesses are asking now.

Law, accounting and consulting run on information clients trust them with. Three duties already in force now reach the AI tools that information passes through.

01 Client confidentiality Solicitors must keep the affairs of current and former clients confidential unless disclosure is required or permitted by law or the client consents (SRA Code of Conduct for Solicitors, paragraph 6.3). In August 2026 the SRA warned that putting client information into AI tools without contractual, technical and organisational safeguards will likely breach that duty. Accountants carry the same duty of confidentiality under their professional ethics codes.

02 UK GDPR and the ICO Sending personal data to an organisation outside the UK, including letting it reach your systems remotely, can be a restricted transfer under UK GDPR. Since February 2026 the test is whether protection there is not materially lower than in the UK. The ICO looks at who you contract with and where they are, so an AI service run from abroad, or a UK supplier with overseas sub-processors, needs checking before client data goes in.

03 Procurement questions Buyers now ask where AI processing happens, not only where data is stored. NCSC guidance tells UK organisations buying cloud services to find out in which countries their data is stored, processed and managed, and which legal jurisdictions it could be subject to. In the public sector, PPN 017 asks suppliers to say whether AI is used in what they provide.

03 · Where your model can run

Where your model can run: two options.

Both can be sovereign when they are set up properly. They differ in who carries the work, how fast you start and how the cost behaves. Neither is right for everyone.

Running AI in your own environment compared with MindLynx hosting it for you
What to compare	Your environmentYour own servers or private cloud	Hosted by MindLynxIn the UK and EU by default
Who owns the hardware	You do, or your cloud provider under your own account.	We do, through named data-centre operators. The security pack lists each one.
Who patches it	Your team: operating systems, drivers and model servers. We supply updates to our software.	We do.
Who holds the keys	You do, in your own key store.	We manage them for the service. The security pack sets out how they are held and who can use them.
Time to first use	Longer. Capacity has to be in place, then installation and testing with your security team, with our engineers alongside yours.	Shorter. Our activation team connects your systems, and every connection starts read-only.
Cost shape	Hardware or reserved capacity paid for up front, plus your people to run it. The cost stays the same when usage falls.	A subscription. Every step has a budget and a price, agreed before it runs and shown after.
Upgrade path	You decide when to move to a new model or version, and you test it first.	We move to newer models. Every receipt names the model that ran, so a change is never hidden.
Audit evidence	Receipts are written inside your environment, into your own logs, where your auditors can read them.	A receipt for every step, which your auditors can read and you can export as plain files.
When a model is retired	Nothing is retired on you. Open-weight models you run stay on your servers until you replace them.	The work moves to another model your rules allow. What your business has learnt stays in Cortex, which is not tied to any one model.

Your own environment is the right answer when

You already run GPUs or a private cloud with spare capacity.
The work is air-gapped or classified and cannot touch a network you do not control.
A regulator, a client contract or your insurer requires it.
Your security team must hold every key and patch every machine.

Hosting with us is the right answer when

You want to start on one problem without buying hardware.
You have no team to run model servers.
Your usage rises and falls through the year.
Processing in the UK and EU, with named operators and a receipt for every step, meets your obligations.

You can start hosted and move into your own environment later. Your knowledge, playbooks and records export as plain files.

04 · We can implement it

Sovereign AI in practice: we set it up with you.

Sovereignty is not only a policy question. It is also a choice about where your AI runs, and who controls it. MindLynx can host Albion for you in the UK and EU, or set it up inside your own environment and train it there on how your firm works. What it learns there belongs to you. Albion is in development, and we are building it with design partners now.

Data privacy In your own environment, your data does not leave. Hosted with us, it stays in the UK and EU by default. It is never used to train anyone else's model.

Security Your rules, never a model, decide what runs and where. In your own environment, you hold the keys and the logs. Every step records who asked, which rules applied, which model ran, where it ran and what it cost.

Cost You do not pay a frontier premium on everyday work. Albion is open source: in your own environment, you pay to run it; hosted with us, every step has a budget and a price. Not every task needs the largest model.

Talk to us about hosting or training Albion See private LLM options

05 · How we're building it

Control first. Then build.

Sovereignty is control, with a British lens. Part of being sovereign is being shaped by national experience, culture and beliefs, so that is how Albion is built: in this order, and in the open.

01 Control first Complete data residency and complete transparency, with data privacy law applied from day one. It will not be perfect, but nothing is lost by starting here, and everything after it inherits it.
02 Start from the models that exist The best open models available today give near-frontier capability from the start. We teach them through our own method: lawful data, models whose licences allow it, and people who teach with consent and fair pay. How people teach Albion →
03 Teach the British lens where it moves the needle Law, finance, health, public services and the institutions behind them: the places where British data, British practice and British culture make the biggest difference to the answer.
04 Train our own, scientifically In parallel, we are training our own model from scratch. We say what we will test before we test it, publish every result, including the ones we lose, and every model card names its base, its teachers and its scores.
05 A receipt for every step Every step records who asked, which rules applied, which model ran, where it ran and what it cost. We build in the open, so you can check the record rather than take our word.

If it takes two years, it takes two years. Because control comes first, nobody is at a disadvantage while the models improve underneath. When our own is ready, it drops into the same place, under the same rules, with the same record.

What it runs on

Albion Albion is our family of AI models, being built now in the UK and Europe and close to going live. It learns only from declared, consented sources, your data is never used to train it unless you say so, and every model card names its base, its teachers and its scores. Explore Albion →

Helix Helix runs the work on your processes, your systems and your rules. Rules, never a model, decide where each piece of work may be stored and run. The law removes places first; only then do cost and speed choose among what is left. If nowhere is allowed, the work is refused and the refusal is recorded. See Helix →

Cortex Cortex is one memory for the whole organisation, held for your business alone. Agents read it before they act and write back what they learn. It is not tied to any one model, so when a model is replaced, what your business knows stays where it is. See Cortex →

06 · Questions

What people ask about sovereign AI.

Need more for your due diligence? Our security pack covers our controls, the places we process data, our subprocessors and how we handle incidents.

Request our security pack

What is sovereign AI?

Sovereign AI is AI where the final say over the model, the data and the computing power rests with the people it serves: the country whose law applies, the company that owns the work, and the individual the information is about. For a business, it means your client files, and the work done on them, stay where you decided, with the evidence to prove it.

What does sovereign mean?

Sovereign means holding supreme or final power. Applied to AI, it asks who has the final say over the model, the data and the computing power: who can read the data, change the model or switch it off.

Who should own AI?

The people it serves, each over its own part: the country whose law applies, the company that owns the work, and the individual the information is about. A business should be able to choose where its work runs and prove it, and a person should keep the right to see, correct and erase what is held about them.

Can we run MindLynx in our own environment?

Yes. Larger organisations can run MindLynx inside their own environment, on their own servers or in a private cloud, with their own keys and logs. It takes longer to set up than our hosted service, and our engineers work alongside yours to do it.

Can MindLynx host Albion for us, or train it in our environment?

Yes. We can host Albion for you in the UK and EU, or set it up inside your own environment and train it there on how your firm works, so your data does not leave. What it learns from your work belongs to you, and your data is never used to train anyone else's model. Albion is in development, and we are building it with design partners now.

Where is our data processed?

On our hosted service, in the UK and EU by default. Every place we use is described by its jurisdiction, who operates it and who holds the keys, and you can narrow it further. In your own environment, it is processed wherever you run it.

Who owns the model weights?

It depends on the model. An open-weight model is published under a licence that says what you may do with it. A closed model's weights stay with its vendor, and you rent access. Every Albion model card names the base it is built from. What Albion learns from your work belongs to you, and your data is never used to train anyone else's model.

Know where your AI runs.

Bring us the tools you use and the rules you answer to. We will show you where the work would run, in your environment or ours, and what the record looks like.

Talk to us about sovereignty Trust & security

Sources