Sovereign AI is a posture,
not a product.

Draft in preparation. A long-form treatment is forthcoming; below is the working outline our editors are circulating.

Working outline

1. The vendor frame is wrong. “Sovereign AI” is being sold as a region, a tenant boundary, or a private model. None of those are sovereignty.
2. Sovereignty is operational. It is who can audit the dataset, who can revoke a fine-tune, who answers a regulator at 3am — not where the GPUs are racked.
3. The posture, in five questions. Data lineage. Model provenance. Inference observability. Human review authority. Incident jurisdiction.
4. What we see in practice. Customers who answered all five on paper still failed two of them in a tabletop exercise.
5. What an honest sovereign stack looks like. A small architecture sketch — not a product diagram.