An air-gapped AI deployment runs the full platform — models, reasoning, audit ledger, operator surfaces — inside a disconnected enclave with no path to external networks. MAIA's sovereign configuration is designed for this: the substrate carries its own models, its policy engine, and its evidence ledger, so classified and disconnected operations keep the same capabilities as connected ones.
The honest question: what breaks offline?
Most AI platforms are thin clients over someone else's cloud: remove the internet and what remains is a login page. Air-gap readiness is decided at architecture time — model hosting, licence checks, update mechanisms, and telemetry all have to be designed to live inside the wire.
MAIA's sovereign configuration was built for exactly this posture: forward-deployed, classified-ready, with in-enclave model serving and updates delivered as verified artefacts rather than live connections.
Evidence matters more in the dark
Disconnected environments amplify the audit requirement — there is no vendor to subpoena later. Every decision inside the enclave seals into the same hash-linked ledger, so an inspector can replay months of operations from the enclave's own records.