Sovereign AI means more than keeping data in-country

Keeping AI data in-country is not much of a sovereignty plan if a business cannot change its model, hosting or support team without starting over. A new Forbes Council essay offers buyers practical checks for finding out how much control they actually have.

Watch Desk analysis

What happened

Hao Sheng, CEO of Expertise AI, argues that buyers should trace a customer interaction from the original request through retrieved documents, the model’s response, backups and support access. He recommends asking which organisations and jurisdictions are involved, who controls access and encryption keys, and who can approve changes to the hosting region or model provider.

Sheng also advises testing the specific model-and-hosting combination a buyer intends to use. To test whether a system can move, his suggested exercises include running a representative task on a second approved model, exporting company knowledge and configuration, and asking for the costs and disruption involved in changing hosting environments. He makes the case in a Forbes Business Council essay published on 19 October.

Why it matters

A local server address does not tell a buyer where every part of an AI interaction is processed, who can access its logs or how difficult it would be to switch providers. Those details affect both data handling and a business’s ability to change course.

Sheng’s checklist turns the broad promise of “sovereign AI” into questions a procurement team can put to a vendor and test in a pilot. It also makes clear that private hosting has trade-offs: organisations may take on responsibility for security, updates, infrastructure and specialist skills.

Our read

This is useful advice from a practitioner, not an independent comparison of AI vendors. The best part is its emphasis on demonstrations: a migration plan that exists only in a slide deck is not much of an exit plan.

Buyers should test a real task, check what they can export and ask who can keep the system running if the original implementation team leaves. Sovereignty need not mean building everything yourself. It should mean knowing which dependencies you have chosen, and what changing them would involve.

What to watch

• Whether vendors can map where data is processed, stored and accessed across a complete interaction.
• Whether pilots demonstrate a second approved model and a usable export of business knowledge and configuration.
• Whether migration costs, support responsibilities and ownership of developed work are clear before a contract is signed.

Discussion spark: Should AI buyers treat the ability to switch models and hosting providers as a basic requirement, even if that adds cost and complexity?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.