· the problem
You do not lose control all at once.
Organisations lose sovereignty progressively, through reasonable decisions taken without an overall vision: integration of a tool, automation of a process, adoption of a model, data transfer, vendor contract, workflow standardisation.
· typical trajectory
weak dependency at first
usage becomes embedded
teams adapt, data flows
exit costs increase
· point of no return
vendor becomes indispensable
exit is prohibitive
data exposed without return
it is no longer dependency — it is capture
Sovereignty is not measured by a vendor's nationality.
It is measured by three questions: can we exit? In what timeframe? At what cost?
· the five levels of ai sovereignty
What must be evaluated and steered.
Infrastructure — mastery of hosting, computing, cloud, storage and access environments
Data — mastery of proprietary data, its usage, circulation and contribution to third-party models
Models — capacity to understand, evaluate, substitute, benchmark or frame the models used
Processes — capacity to maintain a critical process in case of contractual breach or vendor condition change
Decisions — capacity to explain, contest and take back control of AI-influenced or AI-produced decisions
Agents — capacity to define, limit and revoke the rights of action of AI agents operating in your systems
· the essential distinction
An organisation can depend on an LLM, a cloud or an AI vendor — provided that dependency is chosen, steered and reversible.
Dependency is not a problem in itself. Suffered, invisible or irreversible dependency is.
· what you get
A dependency map. A sovereignty framework.
→ complete mapping of existing AI dependencies
→ evaluation of real exit costs for each critical dependency
→ identification of acceptable dependencies vs exposures to reduce
→ sovereignty steering framework (KPIs, milestones, reviews)
→ contractual and technical reversibility recommendations
→ sovereignty positioning relative to competitors
· d-c-a-r applied to ai sovereignty
The four dimensions of a sovereign position.
D
Dependency
Which vendors, models or infrastructures do you depend on? Is the exit cost known and documented?
C
Control
Can you migrate or exit a vendor within a timeline compatible with your strategy? Who decides that exit?
A
Advantage
Do your dependencies create capturable value for you — or do they transfer advantage to your vendors?
R
Responsibility
Do you know who answers if a vendor fails, changes conditions or exposes your data?