Onyx Digital Intelligence.

Air Gaps Are Not Absolute. The Physics Does Not Care.

An air gap is a security measure in which a computer or network is physically or logically isolated from other networks, particularly the internet and untrusted systems. It has no direct network connection through which data can routinely enter or leave. Information transfer, when required, must take place through a separately controlled mechanism, such as removable media or an authorised transfer process.

Air gaps are sold as absolute isolation. No network path, therefore no remote attack path. That framing is incomplete.

An air gap removes the conventional network. It does not remove physics, people, or removable media. Attackers who need to cross one treat those three as the actual attack surface.

Two separate problems are routinely conflated.

Getting malware into an isolated system is not the same operation as getting information out of it. An attacker can solve the first with a USB stick and the second with an electromagnetic, acoustic, optical or thermal channel. Demonstrating that data can leave does not prove that the system can be remotely infected in the first place.

The dominant operational path remains removable media.

ESET’s review of seventeen publicly documented air-gap malware frameworks found the same pattern every time: all seventeen used USB drives as the physical transmission medium for transferring data into or out of air-gapped environments. No confirmed operational use of acoustic or electromagnetic covert channels appeared in that set. The frameworks were built for espionage, targeted Windows, and relied heavily on malicious LNK files or autorun behaviour.

Stuxnet is the clearest historical example. It was most likely introduced into the Natanz environment through removable media, then propagated within the isolated network.

GoldenJackal, analysed by ESET in 2024, demonstrates that the architecture remains viable. In its 2019 campaign, ESET documented a USB-based system for transferring tools into an isolated environment and collecting information from it. The initial compromise vector was unknown; ESET’s reconstruction assumes that a connected machine was already compromised.

This is not exotic. It is operational tradecraft refined over more than a decade.

Laboratory side-channels prove the physics can be forced.

Once code is running on an air-gapped machine, the hardware itself can be turned into a transmitter. Research has demonstrated this across multiple physical media.

Electromagnetic emissions from video cards (AirHopper), memory buses (RAMBO and related work), and USB data buses (USBee) can be shaped into radio signals readable by a nearby receiver. Power-line fluctuations (PowerHammer) and USB power-state changes (SPECTER) offer conducted alternatives. SPECTER manipulates USB power behaviour rather than ordinary USB data traffic; its reported channels and rates are experimental results from an arXiv preprint.

Optical channels include hard-drive activity LEDs (LED-it-GO), keyboard LEDs, screen brightness, and infrared illumination from surveillance cameras (aIR-Jumper). Acoustic channels include cooling fans (Fansmitter), hard-drive actuator noise (DiskFiltration), and power-supply components (POWER-SUPPLaY). Thermal channels (BitWhisper) allow extremely slow communication between adjacent machines by modulating temperature. Vibration-based channels (AiR-ViBeR) use fan-induced movement detectable by a nearby smartphone accelerometer.

These techniques establish feasibility under controlled conditions. Most remain laboratory demonstrations. Bit rates are often low, ranges short, and signal quality fragile. That does not make them irrelevant. It makes them specialised tools for environments where conventional network paths do not exist.

What the evidence actually shows

Infiltration has been demonstrated through removable media and can also exploit maintenance processes, trusted personnel, and compromised components or software introduced through the supply chain. Among publicly documented operational air-gap malware frameworks, removable media has been the consistently observed transmission mechanism. Its practical advantages include comparatively high bandwidth, reliability, and the ability to transfer substantial quantities of information without specialised external receivers. This pattern suggests that operational attackers favour established physical transfer mechanisms over the more specialised covert channels demonstrated in laboratory research.

An air gap raises the cost of an attack. It does not make the attack impossible. It changes the required capabilities from remote network exploitation to physical or supply-chain access plus patience.

Defensive implications

Treat the air gap as one control among several, not as a binary guarantee.

Control removable media with technical enforcement where possible and with process discipline where it is not. Monitor for unexpected USB activity and for the behavioural signatures of known air-gap frameworks. Assume that any system that ever touches an external drive is potentially bridgeable. For the highest-value targets, treat physical access, maintenance procedures, and the integrity of the supply chain with the same seriousness as the network boundary.

Network monitoring alone cannot detect a channel that never touches a network interface. High-assurance environments need broader visibility into physical emanations, unusual hardware behaviour, and environmental conditions.

The physics of the machine does not care about the policy that declared it isolated. Adjacent to secure is not secure.

Sources and scope. ESET, "Jumping the air gap: 15 years of nation-state effort", 2021. ESET, "Mind the (air) gap: GoldenJackal gooses government guardrails", October 2024. Fraunhofer SIT, Stuxnet technical analysis. Mordechai Guri et al.: AirHopper (2014), BitWhisper (2015), LED-it-GO (2017), DiskFiltration (2017), PowerHammer (2018), aIR-Jumper (2019), Fansmitter (2020), POWER-SUPPLaY (2020), AiR-ViBeR (2021), and RAMBO (2024). SPECTER, Choi et al., arXiv preprint, 2026 (USB power-state modulation). Park et al., "A Survey on Air-Gap Attacks", Sensors, 2023. Australian Cyber Security Centre guidance on cross-domain solutions and air-gapped environments.

This piece summarises publicly documented operational frameworks and peer-reviewed or pre-print research demonstrations. Laboratory techniques are not treated as evidence of operational deployment. Reported transmission rates and distances are experimental results under specified conditions, not guaranteed real-world performance. The initial delivery route of Stuxnet and the initial compromise vector of GoldenJackal remain incompletely established in the public record; the article uses the language of the original analyses. Interpretation of relative risk and defensive priority is mine.

Clayton Bax

Published under ONYX Digital Intelligence Following the #OnyxAudit methodology.

"Adjacent to true is not true."

Truth has no flag nor favour, only a standard. And it's heavy!

Editors note: I use various AI agents to fetch my research. I check, strip and write the article. Facts are partly theirs. Interpretation is mine.

#GoldenJackal #South Africa #Stuxnet #USB #air-gap #cybersecurity #privacy #research #side-channel