My collection of covert channels, side-channel attacks, and air-gap exfiltration techniques, organized by physical vector and sensing modality.
Data exfiltration via controlled heat emissions between adjacent air-gapped computers.
Variant of BitWhisper targeting smartphone temperature sensors as receivers, crossing the air-gap via controlled thermal fluctuations from the compromised PC.
Recovery of keyboard input using residual heat patterns captured by thermal cameras.
Thermal signaling detectable at distances >= 2 meters, exploiting environmental heat propagation.
Thermally induced clock drift used as a low-bandwidth covert signaling channel.
Recovery of PINs and keystrokes using thermal imaging of recently touched surfaces.
Modulation of fan speed to generate acoustic signals carrying sensitive data.
Modulation of GPU fan speed for data transmission, extending the Fansmitter concept to discrete graphics cards.
Acoustic data exfiltration from speakerless air-gapped computers via controlled hard-drive seek noise.
Covert acoustic signals generated by manipulating the CD/DVD drive motor and head movement in audioless computers.
Turning power supplies into speakers by modulating load-induced coil vibrations.
Exfiltration using the internal motherboard buzzer (BIOS beep speaker) to transmit ultrasonic signals at 20 bits/sec over 1.5 meters.
Inaudible ultrasonic communication between speakers and microphones.
Acoustic exfiltration through noise generated by rapidly shifting bitmap patterns on LCD screens. Works without any audio hardware (speakers, headphones).
- https://www.darkreading.com/ics-ot-security/air-gapped-networks-vulnerable-to-acoustic-attack-via-lcd-screens
- https://www.instagram.com/reel/C_0Osn6JwHJ/
First attack leveraging smartwatches as receivers for ultrasonic covert communication in air-gapped environments (18-22 kHz). Effective at up to 8 meters with zero bit error rate at 5 bps. Published at IEEE COMPSAC 2025.
Data transmission through mechanical vibrations propagating across solid surfaces.
Modulates hard disk vibrations to encode sensitive data, then uses a COTS mmWave radar to decode from over 20 meters away. Works in NLOS scenarios and with unknown target locations. Published at USENIX Security 2025.
Injecting data into MEMS gyroscopes using acoustic interference.
Exploitation of structural resonance frequencies for improved vibration-based signaling.
Transmission of vibrational signals through the human body to wearables.
Neural decoding of ultrasonic emissions in noisy environments.
A practical deep learning-based attack that identifies keystrokes by sound with very high accuracy.
Audio recovery by measuring laser reflections from vibrating surfaces.
Classic demonstration of laser-based eavesdropping by reflecting a beam off a vibrating surface to reconstruct conversations.
Passive recovery of sound from video of vibrating objects.
Data exfiltration by modulating the hard drive activity LED at speeds invisible to the human eye.
Exfiltration of data by controlling network interface card LEDs to transmit encoded Morse signals.
Controls status LEDs on network switches and routers to transmit covert data at 10 bit/sec to over 1 Kbit/sec per LED.
Uses keyboard indicator LEDs (Caps Lock, Num Lock, Scroll Lock) as an optical covert channel.
Bidirectional air-gap communication using infrared LEDs of surveillance cameras. Exfiltration at 20 bit/sec and infiltration at over 100 bit/sec, reaching hundreds of meters with IR.
Imperceptible modulation of LCD screen brightness to transmit data optically to a nearby camera or light sensor.
Very low contrast LCD display flickering, invisible to the human eye, used to exfiltrate data from air-gapped systems.
Extraction of cryptographic keys by analyzing video recordings of a device's power indicator LED.
Recovery of audio signals from power indicator LEDs.
Data leakage via microscopic reflections of screen content on nearby objects.
Reconstruction of degraded optical signals using deep learning.
Covert RF transmission via USB data bus electromagnetic leakage.
Radio signal generation via SATA cable electromagnetic radiation at 6 GHz.
Using Ethernet cables as antennas for RF data exfiltration.
Software-only generation of Wi-Fi-compatible signals from DDR SDRAM memory buses without any wireless hardware.
Generates EM emissions at cellular frequencies by invoking specific memory-related instructions and utilizing multichannel memory architecture. Received by a rootkit in the baseband firmware of a nearby phone.
FM radio emissions generated by the computer's video card, decoded by the FM radio receiver in a nearby mobile phone. Effective at 1-7 meters.
Covert channel between air-gapped systems and nearby smartphones via CPU-generated magnetic fields detected by the magnetometer.
Malware manipulates memory bus operations to generate controlled radio signals from RAM. An attacker intercepts with an SDR and antenna at up to 1000 bits/sec. Published at NordSec 2024.
Manipulates EM radiation from VGA/HDMI video cables to generate LoRa-protocol-compatible packets. Received by widely deployed COTS LoRa gateways at up to 87.5m (COTS) or 132m (SDR), penetrating multiple concrete walls. Published at ACM CCS 2025.
Keylogging attack on air-gapped computers using electromagnetic emanations without any additional hardware implant.
Recovery of keystrokes via EM side channels from wired keyboard cables.
High-bandwidth EM emissions from PCIe interconnects.
Low-frequency magnetic signaling capable of bypassing Faraday cages.
Data transmission via fluctuations on electrical power lines.
Information leakage via voltage regulator modules.
Data exfiltration using barometric pressure sensors.
Light modulation captured by environmental light sensors.
Data-dependent vibration patterns from storage devices.
Memory disturbance used for covert communication between isolated domains.
3D reconstruction of environments using compromised cameras.
Extraction of information via involuntary human micro-movements.
Modern classification and comparison of air-gap attack vectors.
Review of covert exfiltration techniques with a new clause proposal for ISO 27001 standardization.
- Artificial intelligence significantly increases channel reliability.
- Air-gapping is a mitigation, not a guarantee.
- Physical isolation remains the only robust defense.
- Techniques marked (2024) and (2025) represent the most recent research.
Threat Modeling for Air-Gapped & Side-Channel Attacks
Practical threat assessment mapping air-gap eavesdropping and covert-channel techniques to
real-world environments, focusing on feasibility, impact, and operational risk.
1. Industrial Control Systems (ICS / SCADA)
Environment Characteristics
High-Risk Attack Vectors
Electromagnetic (EM) Channels
Why effective:
Impact:
Power Line-Based Channels
Why effective:
Impact:
Acoustic & Vibration Channels
Why effective:
Impact:
Defensive Reality Check
Overall Risk Level: HIGH
2. Military & Intelligence Facilities
Environment Characteristics
High-Risk Attack Vectors
Magnetic Field Channels
Why effective:
Impact:
Optical Channels
Why effective:
Impact:
Thermal Channels
Why effective:
Impact:
Defensive Reality Check
Overall Risk Level: MEDIUM-HIGH (for targeted adversaries)
3. Hospital & Medical Environments
Environment Characteristics
High-Risk Attack Vectors
Acoustic & Ultrasonic Channels
Why effective:
Impact:
Sensor-Based Channels
Why effective:
Impact:
Optical Channels
Why effective:
Impact:
Defensive Reality Check
Overall Risk Level: HIGH
4. Comparative Risk Summary
5. Key Takeaways
Physical isolation is the only robust mitigation -- and is rarely absolute.
6. What Defenders Think vs. Reality
Assumption 1: "Air-gapped means no data can leave"
What defenders think:
Removing network connectivity eliminates exfiltration paths.
Reality:
Air-gaps only remove logical channels.
Physical channels (EM, acoustic, optical, thermal, magnetic) remain fully exploitable.
Consequence:
Security posture is built on a false binary model: networked vs isolated.
Assumption 2: "Low bandwidth attacks are irrelevant"
What defenders think:
If an attack only leaks bits per second, it is not operationally useful.
Reality:
All require very little bandwidth.
Consequence:
Low-bandwidth channels are ideal for long-term espionage.
Assumption 3: "Environmental noise makes attacks impractical"
What defenders think:
Noise (EM, acoustic, thermal) masks covert signals.
Reality:
Consequence:
Noise increases stealth, not safety.
Assumption 4: "TEMPEST compliance covers these threats"
What defenders think:
TEMPEST standards mitigate EM leakage risks.
Reality:
Consequence:
Compliance != security.
Assumption 5: "If it were real, we would see it in the wild"
What defenders think:
Lack of public incidents implies low risk.
Reality:
Consequence:
Threat modeling lags behind attacker capability.
7. Countermeasures by Environment
7.1 Industrial Control Systems (ICS / SCADA)
Practical Countermeasures
Illusory Countermeasures
7.2 Military & Intelligence Environments
Practical Countermeasures
Illusory Countermeasures
7.3 Hospital & Medical Environments
Practical Countermeasures
Illusory Countermeasures
8. Final Reality Check
If physical effects exist, they can be weaponized.
If they can be weaponized, they eventually will be.
9. Exfiltration vs Passive Leakage
Does Air-Gapped Exfiltration Require Prior Contamination?
9.1 Definitions
Prior Contamination
A system is considered contaminated if any attacker-influenced logic executes, including:
A system with no prior contamination behaves exactly as designed.
Exfiltration
True exfiltration must be:
Anything that does not meet these criteria is not exfiltration.
9.2 Core Conclusion
What is possible without contamination is passive side-channel leakage.
9.3 Cases That Do NOT Require Prior Contamination (Passive Leakage Only)
The following cases observe unavoidable physical emissions produced during normal operation.
They do not encode data and cannot be used to extract arbitrary information.
9.3.1 Electromagnetic (EM) Emanations
Examples: EM leakage from wired keyboards, classical TEMPEST-style eavesdropping
Why no contamination is needed: Keystrokes naturally modulate EM fields
What can be recovered: Keystrokes, timing information
Hard limitations: Only during active typing. No access to stored data. No persistence.
9.3.2 Optical Leakage
Examples: Power / HDD / network LEDs, screen reflections, visual microphone scenarios
Why no contamination is needed: LEDs and displays expose activity by design
What can be recovered: Activity patterns, limited cryptographic operations during execution
Hard limitations: No encoding control. Highly context-dependent. Line-of-sight required.
9.3.3 Acoustic Leakage from Human Interaction
Examples: Typing sounds, mechanical keyboard clicks, mouse usage
Why no contamination is needed: Sounds are unavoidable byproducts of use
What can be recovered: Keystroke inference, behavioral patterns
Hard limitations: Environmental noise. No bulk data leakage.
9.3.4 Thermal Residue Leakage
Examples: PIN recovery from keyboards, touchscreen thermal imaging
Why no contamination is needed: Heat is a natural physical residue
What can be recovered: Recent inputs only
Hard limitations: Short time window. No persistence. No stored data access.
9.3.5 Human-Mediated Leakage
Examples: Shoulder surfing, CCTV observation, reflections on glasses or glossy surfaces
Why no contamination is needed: Information is voluntarily displayed to humans
Hard limitations: No automation. No scalability. No long-term channel.
9.4 Important Non-Cases (Often Misclassified)
Sensor Injection Attacks
Examples: Gyroscope injection, microphone injection
Clarification: These attacks inject signals into the system. They do not extract data from the system.
-> Not exfiltration.
"AI Enables Exfiltration Without Contamination"
False.
AI improves: signal reconstruction, noise filtering, decoding accuracy.
AI does not: select data, encode bits, control timing.
-> Encoding still requires code execution.
9.5 Why Exfiltration Fundamentally Requires Contamination
To exfiltrate arbitrary data, an attacker must:
No passive physical phenomenon provides these capabilities.
9.6 Taxonomy Summary
9.7 Correct Threat-Model Statement
9.8 Integrated Takeaways
Air-gapping mitigates risk -- it does not provide absolute isolation.
10. Alignment with NIST SP 800-53 and MITRE ATT&CK for ICS
10.1 NIST SP 800-53 Perspective: What Air-Gaps Actually Control
Under NIST SP 800-53, air-gapping primarily supports controls related to
information flow restriction, not absolute data protection.
Relevant Control Families
AC - Access Control
Air-gapping enforces logical information flow restrictions.
It does not prevent physical or side-channel information flows.
IA - Identification and Authentication
Authentication controls are out of scope for side-channel exfiltration.
SC - System and Communications Protection
SC controls largely assume network-mediated communication and explicit transmission channels.
Side-channel exfiltration operates outside the threat model assumed by SC controls.
PE - Physical and Environmental Protection
Most organizations implement PE controls for theft, tampering, and unauthorized access.
They rarely implement PE controls for EM emissions, optical leakage, or acoustic/thermal channels.
NIST-Compliant Statement
10.2 NIST View on Prior Contamination
From a NIST standpoint, prior contamination maps to failures in multiple control families:
Air-gaps do not mitigate failures in SR (Supply Chain Risk Management) or MA (Maintenance).
10.3 MITRE ATT&CK for ICS: Where Exfiltration Fits -- and Where It Doesn't
MITRE ATT&CK for ICS is explicitly scoped to observable adversary behaviors,
logical and operational attack paths, and networked/host-based activity.
Relevant Tactics
Collection: ATT&CK focuses on logical data collection. Side-channel leakage during use is out of scope.
Exfiltration: Techniques assume network protocols, removable media, or logical channels.
Physical side-channel exfiltration is largely outside the current ATT&CK for ICS taxonomy.
Why This Matters
10.4 Correctly Positioning Side Channels in ATT&CK-Based Threat Models
To remain ATT&CK-consistent, side-channel attacks should be modeled as:
Recommended Language
10.5 Mapping Prior Contamination to ATT&CK for ICS
ATT&CK models how access is obtained, not how physics is abused.
10.6 NIST and ATT&CK