Air-gapping is the gold standard of network isolation. By physically disconnecting a high-security computer from the internet and local networks, organizations assume that sensitive data cannot escape. Traditional network security models operate on the assumption that data transfer requires a dedicated communication interface—a Wi-Fi card, an Ethernet cable, or a cellular modem.
However, computer hardware is fundamentally physical. Every digital operation requires moving electrical charges, which inherently dissipates heat, draws current, generates magnetic fields, and toggles status indicators. A growing body of research demonstrates that malware can programmatically manipulate these standard, unprivileged hardware components to translate digital data into structured physical emissions. By modulating these emissions, an attacker can turn an isolated machine into an unintended transmitter, bridging the air-gap without any specialized hardware modifications.
The Core Intuition: Workload as a Modulated Carrier
To understand why these attacks work, consider how a radio transmitter operates: it generates a high-frequency carrier wave and alters its amplitude, frequency, or phase to encode binary information.
In an air-gapped system, malware cannot access a radio antenna. Instead, it uses the system's core components—like the CPU or GPU—as a proxy. When a processor executes an intensive task (such as a tight loop of floating-point operations), it draws a significant amount of electrical current. When it idles or sleeps, its current draw drops precipitously.
By rapidly toggling between these two states, software can intentionally shape the machine's aggregate physical footprint over time. If the malware switches states at a specific frequency, it creates a physical "square wave" that propagates through the environment. A nearby receiver can then capture this signal and decode it back into binary data.
Software Domain Physical Domain Receiver Domain
--------------- --------------- ---------------
Data: [ 1 ] ---> CPU Stress ---> High Power / High Emission ---> Detected Peak ---> Data: [ 1 ]
Data: [ 0 ] ---> CPU Idle ---> Low Power / Low Emission ---> Detected Trough ---> Data: [ 0 ]
Modulating Power Lines and Magnetic Fields
The most direct consequence of workload manipulation is its effect on the system's power infrastructure. The fluctuations in current draw do not stay confined to the silicon; they travel backward into the power supply unit (PSU) and out through the electrical wiring.
In PowerHammer, Mordechai Guri et al. demonstrated that programmatically toggling intensive CPU tasks allows malware to inject readable data patterns directly into a computer's physical power cord. Similarly, Zhihui Shao et al. showed in Your Noise, My Signal that this workload modulation creates distinct switching noise in standard desktop power supplies, allowing data to be intercepted further down the building's power grid. This physical-layer leakage bypasses software-defined security boundaries entirely. For instance, Riccardo Spolaor et al. developed an attack called PowerSnitch in No Free Charge Theorem, proving that even when a mobile device is connected to a power-only USB charging cable with its data pins physically disconnected, sensitive data can still be serialized and leaked through the remaining power lines.
According to Ampere's Law, every electrical current also generates a proportional magnetic field. When malware modulates the current flowing through CPU cores, it alters the surrounding magnetic field. Yuval Elovici et al. leveraged this phenomenon in MAGNETO, showing that CPU-generated magnetic fields can be picked up by the magnetometers built into standard smartphones. Because low-frequency magnetic fields easily penetrate conductive materials, the authors of ODINI demonstrated that this mechanism can even escape highly secured environments protected by Faraday cages, which typically block conventional radio frequencies.
Software-Defined Radio via High-Frequency Buses
While power line and magnetic attacks rely on low-frequency baseband modulation, malware can also generate higher-frequency electromagnetic radiation by exploiting internal data buses. Computer buses and video cables act as accidental antennas; when high-speed digital signals pass through them, they emit electromagnetic waves.
In AirHopper, Mordechai Guri et al. showed that by displaying specific, carefully crafted pixel patterns on a standard monitor, malware can force a video display adapter to radiate electromagnetic signals in the FM radio band. This effectively transforms the graphics hardware into an FM transmitter capable of beaming passwords to a nearby smartphone's radio receiver.
This concept applies to other hardware interfaces as well. In USBee, Mordechai Guri et al. demonstrated that writing specific bit sequences to an ordinary, unmodified USB thumb drive generates controlled electromagnetic emissions from the USB bus itself, creating a short-range radio transmitter out of a storage device without requiring administrative privileges.
As the sophistication of these software-defined transmitters has grown, their range has expanded significantly. Xieyang Sun et al. introduced TEMPEST-LoRa, a technique that manipulates video cable emissions so precisely that they mimic the exact packet structure of the LoRa (Long Range) wireless protocol. This cross-technology communication allows an air-gapped computer to transmit data directly to an off-the-shelf LoRa gateway nearly 90 meters away, penetrating concrete walls without any native wireless hardware.
Exploiting Optical and Periphery Indicators
Beyond invisible electromagnetic and electrical fields, attackers can exploit the visible spectrum by controlling status indicators on peripheral devices. Many systems feature LEDs (such as keyboard caps-lock lights or hard drive activity indicators) that are programmatically controllable via standard operating system APIs or low-level Human Interface Device (HID) packets.
In CTRL-ALT-LED, Mordechai Guri et al. demonstrated that malware can bypass high-level API monitoring by using low-level HID USB packets to toggle keyboard status LEDs at thousands of bits per second. While these rapid fluctuations are invisible or appear as a steady glow to the human eye, they can be captured by nearby security cameras or optical sensors. Zheng Zhou et al. expanded on this vector in Optical Exfiltration of Data via Keyboard LED Status Indicators to IP Cameras, highlighting that static status indicators are frequently overlooked during security audits despite being highly vulnerable to programmatic modulation.
[Malware] ---> HID USB Packets ---> Keyboard LED (High-Freq Blinking) ---> [IP Camera] ---> [Attacker]
Optical channels can also achieve remarkable bandwidth and range. Niclas Kühnapfel et al. developed LaserShark, a system that repurposes built-in device LEDs to establish fast, bidirectional optical communication over distances of up to 30 meters, reaching speeds of 100 kbps. Furthermore, the integration of smart environments introduces new vulnerabilities; Jieun Lee 0002 et al. noted in Optical Air-Gap Attacks that internet-connected IoT components like smart bulbs provide large, controllable light-emitting surfaces that can be co-opted as data transmitters.
In some cases, the peripheral itself can be weaponized to interact with nearby consumer electronics. Zheng Zhou et al. demonstrated in Potential risk of IoT device supporting IR remote control that a modified keyboard can emit infrared signals that mimic standard remote control operations. This allows an air-gapped system to proxy data through an unwitting, internet-connected smart TV or appliance without needing to compromise the IoT device itself.
Thermal and Acoustic Channels
When electromagnetic or optical lines of sight are completely unavailable, malware can fall back on mechanical and thermodynamic properties.
Because computing hardware generates heat as a byproduct of electrical resistance, sustained workloads raise a component's temperature. In BitWhisper, Mordechai Guri et al. showed that two adjacent computers can establish a bidirectional thermal link. By using controlled spikes in CPU and GPU workloads to generate heat patterns, one system can transmit binary data that is subsequently detected by the built-in thermal sensors of the neighboring machine.
Mechanical components can similarly be forced to emit acoustic signals. In Acoustic Data Exfiltration from Speakerless Air-Gapped Computers via Covert Hard-Drive Noise ('DiskFiltration'), Mordechai Guri et al. proved that even if a computer lacks internal speakers or audio hardware, malware can control the physical movements of a hard disk drive's (HDD) actuator arm. By forcing the drive to perform specific seek operations, the software generates distinct acoustic frequencies that can be recorded by a nearby microphone or smartphone, turning mechanical friction into a reliable data stream.
Summary
The unifying insight across this domain is that isolation is a multi-layered physical problem, not just a logical one. As long as software can control the timing and intensity of hardware execution, it can manipulate the physical environment. By treating power lines, magnetic fields, radio emissions, status lights, heat, and acoustic vibrations as programmable serialization mediums, software-defined physical covert channels demonstrate that absolute isolation requires monitoring the physical emanations of computing itself.
Key Foundational Work
AirHopper: Bridging the Air-Gap between Isolated Networks and Mobile Phones using Radio Frequencies (2014) — contributed 3 concept elements to this direction.
Protecting Confidential Data on Personal Computers with Storage Capsules (2009) — contributed 1 concept elements to this direction.