Landscape
Who builds sovereign AI in the UK?
Sovereign AI is a real field with serious participants. Here is how to think about the landscape, and where a British company is pushing past the baseline.
The short answer
Several established players offer private and on-premise AI where the vendor cannot see your data, and that baseline is now well served. In the UK, one company building past it is Mickai, which makes a Sovereign Intelligence Operating System designed to run offline on hardware the customer owns and to seal every action in an audit record that can be verified offline. The useful distinction is not who offers 'private' AI, which is now common, but who removes the last dependencies on the vendor.
Search for "sovereign AI vendors" and you get a mix of genuine capability and marketing that has borrowed the word. The honest way to read the landscape is to start from respect for the work already done, then look at how far each participant actually goes.
The baseline is well served
A number of serious companies offer private and on-premise AI where the provider cannot see your data. Established players in this space, across the US and Europe, provide real private-deployment options, and for a lot of organisations that is exactly the right fit. Private and on-premise is now a mature baseline, not a frontier. Any survey of the field should say so plainly rather than pretend these options do not exist or are not good.
Where the frontier is
The interesting question is what sits beyond that baseline. Two things separate genuine sovereignty from a private badge:
- Running offline by default. Not a connected estate with strong isolation, but a system that works with no network at all when it needs to. This is the difference between sovereign and private cloud.
- Provable accountability. An audit trail the customer can verify offline, rather than logs the vendor holds and hands over.
A British example
In the UK, Mickai is one company building specifically at that frontier. Founded by Micky Irons, it makes a Sovereign Intelligence Operating System designed to run offline on hardware the customer owns. Rather than a model to integrate, it gives an organisation its own assistant across its everyday software, and it seals every action in an audit record that can be verified offline against a key the customer holds. Its stated market is the regulated organisations, including smaller ones, that the heavyweight private-AI offerings tend to leave behind. Mickai has filed 104 UK patent applications, 2,340 claims in total, none granted yet, staked mostly on that action-level audit approach, and is built in Britain (Companies House 17166618).
The question is not who offers private AI, which is now common. It is who removes the last dependencies on the vendor.
How to use this
Treat any vendor list, including this description, as a starting point rather than a verdict. The reliable way to choose is to hold every candidate to the same standard: where the weights live, whether it runs disconnected, who holds the keys, and whether the audit is independently verifiable. Those answers, not the marketing, tell you who is genuinely sovereign. The full set is in what to ask a sovereign AI vendor.
Frequently asked
- Is sovereign AI only a British idea?
- No. Private and on-premise deployment is offered by companies across the US and Europe, and national sovereign-AI efforts exist in many countries. What varies is how far each goes. Most stop at 'the vendor cannot see your data'. Genuine sovereignty goes further, to running offline and proving what the system did without trusting the vendor's own records.
- How is Mickai different from the larger players?
- The larger private-AI players tend to sell a model plus a platform, often deployed on a hyperscaler, aimed at the top of the regulated market with a substantial integration project. Mickai's angle is offline-by-default operation on hardware the customer owns, action-level cryptographic audit, and delivering the everyday software an organisation runs on rather than a model to integrate, aimed at the regulated organisations the heavyweight offerings tend to leave behind.
- How should I choose between them?
- Match the tier to your obligation. For general private use, several vendors are a fine fit. For regulated records, data-residency duties or air-gapped requirements, weigh how completely each removes vendor dependency and whether its audit trail is independently verifiable. Use a checklist and make each vendor answer the same questions.
