Briefs
Sovereignty is an architecture problem
Why data residency promises and contractual controls do not produce the property that regulated buyers actually need.
· 3 min read
[TBD: draft] — This is a placeholder. The outline below records the intended argument. No claim here has been reviewed for publication.
What this brief will cover
Most "sovereign AI" offers are policy instruments. They describe where data will be stored, who may access it, and what the vendor promises not to do. Those are useful commitments, and they are a different kind of thing from a system that cannot send data anywhere.
This brief will draw that distinction precisely, and argue that for classified, controlled, and regulated workloads the architectural property is the only one that survives contact with an accreditation process.
Intended structure
- The difference between a promise and a constraint
- Why default-deny egress changes the review conversation
- What a defined boundary gives an accreditor that a shared service cannot
- Where policy controls remain necessary and sufficient
Status
[TBD: draft] — copy pending review.