Air-Gapped AI

Air-gapped AI means running language models or other AI systems on computers that are physically and logically isolated from any external network — most importantly the public internet. The term borrows from the security concept of the air gap: per the U.S. National Institute of Standards and Technology (NIST/CNSSI) glossary, an air gap is "an interface between two systems at which (a) they are not connected physically and (b) any logical connection is not automated." Applied to AI, it means your prompts, documents, and model weights stay inside hardware you own.

Why sovereign operators care

Most AI today is consumed as a cloud API: your data travels to a third party, lands under their jurisdiction, and is subject to their logging, retention, and the laws of where their servers sit. An air-gapped, on-premise model removes that exposure entirely. For sensitive material — legal files, business strategy, mining operations data, anything you would not paste into a public chatbot — isolation is the strongest privacy guarantee available, because data that physically cannot leave cannot be subpoenaed, breached in transit, or harvested for training.

How it works in practice

Open-weight models can run on a local workstation or server with a capable GPU. The machine is kept off the internet (or behind a strict one-way firewall), updates are moved in manually, and inference happens entirely offline. This is the most demanding tier of self-hosting — it trades convenience for control. For Canadian and Quebec organizations weighing where AI workloads should live, air-gapping is one answer to cross-border legal reach over cloud data.

This is general information, not legal advice — consult a qualified professional for compliance decisions about data handling and AI deployment. See also our guides on Quebec Law 25 and on-premise LLMs and the U.S. CLOUD Act and Canadian AI data.