Sovereignty is process
Why 'national' doesn't mean 'sovereign'
Distinguish a sovereign AI from a national one — and see why co-design is where that distinction is actually decided.
Reading
A common shortcut today is to call a model sovereign because it was trained in-country, on national data, or with public funding. None of those are sufficient. A national model trained from the centre is not, by virtue of being national, accountable to the language communities inside it.
Sovereignty depends on the process: whether each community can shape the social and cultural composition of the data that ends up representing it. That is a question about who sits in the meetings before the dataset is assembled — not about which flag is on the rack-side sticker after deployment.
That makes co-design a sovereignty mechanism, not a courtesy workshop after the roadmap is already fixed. The represented people need authority while the question, sources, labels, tests, refusal paths and retirement conditions are still being designed. A system built for a community, but not with it, can still reproduce the extractive pattern this playbook is trying to interrupt.
Put plainly: the sovereign property of a system is not its origin. It is the answer to "can the people represented inspect, correct, reject, move away from, or retire it?"
“Sovereignty depends on the process. The more important question is whether each language community can shape the social and cultural composition of the data it represents.”
Practise
Exercise
Map the pipeline
- 01Pick one language community you work with or care about. It does not need to be your own.
- 02Sketch the rough pipeline that produces an AI feature touching that community — data collection, labelling, training, evaluation, deployment, feedback.
- 03Beside each stage, write a single word: who decides? If the same name appears three or more times, circle it.
- 04Write down two small co-design interventions that would put community authority into two of those circles. Keep them realistic — a meeting before requirements are frozen, a reviewer named by the community, a contract clause the community can invoke.
Knowledge check
What makes an AI system sovereign, in this playbook's framing?
Which of these is NOT, by Tang's account, a feature of a sovereign system?