What is air-gapped AI?

Definition

An air gap uses physical and network separation to keep models, data and computation inside a controlled boundary. Organizations use this approach when sensitive evidence or regulated information must not leave the local environment.

Isolation changes how data, models and updates enter the system. Transfers need controlled ingestion, validation and auditable hardware boundaries, while local compute limits may require routing work to smaller models and preserving reproducible evidence inside the enclave.

ELI5

Air-gapped AI runs inside a computer environment that has no direct connection to the public internet or less trusted networks. The separation helps keep sensitive data and model activity inside a controlled boundary.

For example, a laboratory might analyze confidential evidence with a local model that cannot send the files to a cloud service. Models, updates and new data must then enter through a controlled process, and the local hardware may limit which models can run.

Acronyms and aliases

offline AI variant

Frequently asked questions

Why run AI in an air-gapped environment?

It reduces the risk that sensitive data will leave the controlled environment through a network connection or external service.

Can an air-gapped AI system receive updates?

Yes, but updates must enter through a controlled, validated and auditable transfer path that does not create an unrestricted return channel.

Videos explaining air-gapped AI