Black boxes and borrowed keys

Dependence on closed, foreign-controlled AI stacks carries costs that rarely appear in the procurement file. They appear later, at the worst possible time.

A red padlock on a computer keyboard
Photo: FlyD on Unsplash

Buying a capable AI service from a large overseas provider is often the fastest route to a working demo. For many uses, it is a perfectly reasonable choice. For systems that underpin public services, security or critical infrastructure, the hidden costs deserve a closer look.

Four dependencies to check

Where the data goes

Sensitive data can end up outside the institution’s control, processed or stored in places and under laws the institution does not govern.

Who holds the keys

Licences, updates and support may depend on a vendor whose priorities can change. A pricing decision or a policy change elsewhere becomes an operational risk at home.

Whether you can see inside

Closed “black-box” models are hard to inspect or adapt. When a recommendation goes wrong, the institution may not be able to explain why, which is a serious problem in courts, audits and parliaments.

What sits underneath

Foreign chips and AI stacks underpin a great deal of critical enterprise infrastructure. Supply chains for that hardware are exposed to forces no single institution can control.

You do not truly own a capability you cannot inspect, adapt or replace.

Your decisionsModels you can inspectKeys you holdInfrastructure you control
Control at every layer: no black boxes, no borrowed keys.

Not isolation: optionality

The answer is not to refuse global technology. It is to architect for optionality: open interfaces, data that stays under local control, models that can be tested against local conditions and the ability to swap components without rebuilding everything. That is what turns dependence back into choice.

Want to explore how this applies to your organisation? We share our approach with qualified organisations in a private briefing.