An air gap stops network packets. It does not stop a digital display cable from radiating electromagnetic energy while it moves billions of signal transitions between a computer and a monitor.

Researchers from Shandong University and Quan Cheng Laboratory turned that leakage into TrojPix, a covert channel presented at USENIX Security 2026. Their prototype changed pixel values by amounts users could not perceive, then decoded the resulting emissions with radio equipment at a distance.

The result deserves attention. It also deserves an honest threat model.

TL;DR

  • TrojPix encodes data through subtle pixel changes that alter electromagnetic emissions from a digital video cable.
  • The researchers reported a peak rate of 8.1 Mbps and recovered a complete packet at a maximum tested distance of 208 metres outdoors.
  • They evaluated 15 commercial cables from eight manufacturers and 12 monitors from nine manufacturers, plus a 30-centimetre concrete-wall test at 10 metres.
  • The attack assumes malware is already running on the air-gapped computer and that the attacker has radio receiving equipment in range.
  • This is mainly a high-assurance, military, government, research, and critical-infrastructure concern—not a reason for ordinary users to wrap HDMI cables in kitchen foil.

The Cable Is the Transmitter

Digital display links carry pixel data as high-speed electrical transitions. TrojPix focuses on Transition-Minimized Differential Signaling, or TMDS, used by HDMI and related digital video paths. Changing a pixel changes the encoded bit pattern. Changing the bit pattern changes the cable’s electromagnetic emissions.

The transmitter is software. TrojPix modifies selected pixel values—during the paper’s feasibility work, the least significant bit of the blue channel—so that the displayed image remains visually almost identical while the emissions carry an attacker-selected signal.

The receiver is physical radio equipment: an antenna, filtering and amplification, and a decoder that recovers the modulated data. The paper does not claim that a laptop across the street can receive the channel with its ordinary Wi-Fi hardware.

Two Operating Modes

The researchers tested two conceptual modes.

Fake screen-off mode displays a black or standby-like screen while the cable continues carrying crafted video data. The malware stops transmission when it detects user interaction and restores the display.

Foreground embedding mode introduces small changes into whatever content is already visible. In a subjective study of 50 volunteers, none reported a visible difference between the original and modified displays used in the test.

That result applies to the researchers’ implementation and test material. It is evidence of perceptual stealth under those conditions, not proof that every screen, user, colour profile, or content type will hide every implementation.

What the Numbers Actually Mean

The headline figures come from different evaluation conditions and should not be combined into a fictional 8.1 Mbps link at 208 metres.

  • The 8.1 Mbps peak was reported with a 20 MHz receiver sampling rate in the sampling-rate evaluation.
  • At 208 metres outdoors, the researchers report recovering one complete packet with synchronization and CRC validation.
  • Between 20 and 120 metres, raw bit-correctness declined from 99.19% to 91.02% as distance increased.
  • In a separate through-wall test, the antenna was 10 metres from the target with a 30-centimetre concrete wall in the path. Average bit correctness declined from 99.96% to 99.14%.
  • A 10 MB test transfer at 10 metres completed in 41.6 seconds, corresponding to about 2.016 Mbps in that experiment.

The distinction matters. “8.1 Mbps and 208 metres” accurately describes the prototype’s separate peak rate and maximum demonstrated range. It does not establish that both maxima were achieved simultaneously.

The Attack Starts After Compromise

TrojPix does not provide initial access. The paper assumes:

  1. Malware has already been implanted on the isolated computer.
  2. The malware can find and read the data it wants to steal.
  3. A digital display cable is active.
  4. The attacker can position suitable receiving equipment within a workable range.
  5. The environment does not attenuate or drown out the useful emissions enough to break reception.

The prototype transmitter ran in user mode and did not require administrator rights or direct peripheral access. That is meaningful after compromise, but it does not explain how malware crossed the air gap in the first place. Removable media, maintenance laptops, malicious insiders, supply-chain access, and unsafe update processes remain the more fundamental controls.

Why Simple Shielding Is Not a Complete Fix

The researchers wrapped an already shielded cable with four additional materials. Tinned copper mesh produced the largest degradation in their test, but the reported success rate remained above 91% for every material evaluated. Their conclusion was that added shielding reduced performance without eliminating the channel.

They discuss stronger physical controls such as Faraday-style shielding, signal interference, and optical display interfaces. Some of those options are expensive, operationally difficult, regulated, or capable of interfering with legitimate equipment. Radio jamming is not a casual recommendation and may be illegal without appropriate authority.

For a genuinely high-assurance environment, the practical defensive stack is broader:

  • prevent and detect unauthorized code execution on isolated systems
  • strictly control removable media, updates, maintenance devices, and software provenance
  • allow only approved applications and display behavior
  • monitor physical access and unexplained radio equipment near protected zones
  • include display hardware and cabling in emission-security assessments
  • use certified shielding, filtered interfaces, optical isolation, or protected zones where the threat model justifies them
  • test controls against the actual room, cable paths, walls, and equipment instead of relying on the word “air-gapped”

What TrojPix Changes

TrojPix does not make every monitor a remotely exploitable radio. It shows that software already running on a target can manipulate a component defenders often exclude from the data-flow diagram: the physical signal leaving the graphics interface.

That is the useful lesson. An air gap is a network control, not a complete information-flow proof. In high-assurance systems, power, sound, light, heat, magnetic fields, and electromagnetic emissions all deserve a place in the threat model—even when the monitor looks perfectly normal.

Sources