Air Gap

An air gap is a complete physical separation between two systems. A computer or an entire network is simply not connected to the internet — no cable, no Wi-Fi, no Bluetooth connection, nothing. Data can only enter or leave by physical means: via USB stick, DVD, or some other storage medium that a person carries by hand. The name describes it literally — between the protected system and the rest of the world lies, figuratively speaking, only air.

Protection through complete isolation

Most attacks on computers occur over network connections. An attacker sends malware via email, exploits a vulnerability in a website, or breaks in through a poorly secured remote maintenance connection. All of these methods require that the target be reachable in some way. An air gap cuts off exactly this attack surface entirely.

That sounds like a simple solution, but it comes at a high price: the system loses almost all convenience. Updates must be installed manually. Files can’t simply be sent by email. Every interaction with the outside world requires a person who is physically present. For most everyday applications this would be absurdly impractical — which is why air gaps remain reserved for certain high-security environments.

How data gets in and out anyway

The only permitted data transfer runs via physical storage media. A technician copies a software update onto a vetted USB stick, carries it into the secured room, and installs it. This procedure is sometimes also called a “sneakernet” — a humorous term alluding to the fact that data is literally transported in sneakers.

Security researchers have shown, however, that even this isolation is not absolute. An infected USB stick that someone unknowingly introduces can transmit malware. Even more sophisticated attacks exploit unexpected channels: sound waves generated by fans, fluctuations in power consumption, or even electromagnetic emissions from a monitor can theoretically transmit tiny amounts of data to the outside. These methods are highly complex and rare in practice — but they show that an air gap drastically raises the security requirements for the physical environment, not just for the software.

The most famous real-world attack on an air-gapped system was Stuxnet, a piece of malware that sabotaged control systems at Iranian uranium enrichment facilities around 2010. Stuxnet made its way into the isolated network via infected USB sticks — demonstrating to the world that an air gap is no cure-all, but only one part of a larger security concept.

Where air gaps are used today

Nuclear power plants, power grids, and other critical infrastructure frequently use air-gapped systems for their control computers. Military command centers and intelligence agencies isolate computers that process particularly sensitive information. Industry also has production facilities whose control systems have never been connected to the internet — often less out of a deliberate security calculation than historical happenstance, because they were built before networking became standard.

In the AI context, the term comes up when companies or governments want to protect particularly sensitive AI models or training data. A model trained on classified government documents must not run on a server reachable from the internet. Some organizations are also discussing air gaps as a security strategy for future, especially powerful AI systems — to prevent such a system from communicating with the outside world in an uncontrolled way.

In the tech press, the term frequently appears in reports about cyberattacks on infrastructure or in debates about AI safety. If you read that a system is “airgapped,” you can assume: it has been deliberately and completely cut off from the internet — and this is not a technical limitation, but a design decision.