Case · DiggingBeagle record

AirHopper turned display hardware into an FM-band exfiltration path

AirHopper research demonstrated data exfiltration from an isolated computer to nearby infected mobile phones using software-generated FM-band electromagnetic emissions associated with display hardware.

Scope

Laboratory covert-channel research requiring compromise of the air-gapped computer and a nearby receiver. It is not evidence of autonomous AI escape.

Assessment profile

Assessment method
Setting
research demonstration
Exploitation
demonstrated
Evidence
Grade B
Basis and provenance

Recorded paper claims describe an FM laboratory demonstration with measured 1–7 m reception, not observed hostile deployment. Assessment is a desk review of the retained canonical Claims and cited Source metadata at their recorded cutoff, not a new external verification. Impact remains ungraded.

Assessed Sep 24, 2026 using diggingbeagle.assessment/1.

At a glance

Mechanism and trust boundary

  1. 01

    Compromise the isolated computer

    Malicious code must already execute on the air-gapped system; AirHopper is an exfiltration path after compromise, not an initial-access technique.

    Boundary: air-gapped host / attacker-controlled code

  2. 02

    Modulate display activity

    The transmitter varies software-controlled display behavior so the display subsystem produces data-dependent electromagnetic emissions.

    Boundary: software activity / display electronics

  3. 03

    Radiate an FM-band carrier

    Video/display hardware and cabling unintentionally act as a physical emitter whose signal can be modulated by the workload.

    Boundary: computer hardware / surrounding electromagnetic environment

  4. 04

    Receive on a nearby mobile device

    A prepared or compromised phone within the tested range captures the emissions and extracts the encoded symbols.

    Boundary: physical emission / nearby receiver

  5. 05

    Forward recovered data

    Once the phone has decoded the secret, ordinary connectivity on the receiver side can carry it beyond the isolated room.

    Boundary: local covert channel / external network

Claims & evidence

CLM-AIRHOPPER-RANGEThe paper reports exfiltration to mobile phones at 1-7 meters with effective bandwidth of 13-60 bytes per second in its tested configurations.supported

Basis: reported finding

Link to claim
CLM-AIRHOPPER-MECHANISMAirHopper demonstrated an FM-band covert channel from an isolated computer to nearby mobile phones without using an ordinary network connection.supported

Basis: reported finding

Link to claim

Implications

AirHopper invalidates the narrow assumption that removing network interfaces removes every communication path. The demonstrated risk is local, receiver-dependent exfiltration from an already-compromised host. The broader defensive implication is that display hardware, cables and nearby radios belong inside a high-assurance containment model when software can modulate them.

Controls and mitigations

  • Keep untrusted mobile and radio-capable devices outside the physical security zone so the demonstrated receiver path is unavailable.
  • Use RF monitoring and shielding appropriate to the frequencies and equipment in the protected environment.
  • Constrain untrusted code execution and unusual high-rate display manipulation on systems whose emanations matter to the threat model.

Unknowns and contradictions

  • The published result does not establish reliable operation across arbitrary modern display interfaces, cables, chassis or shielding designs.
  • Range and throughput depend on the emitting hardware, receiver, geometry and electromagnetic noise.
  • No cited evidence shows an AI system autonomously discovering or deploying AirHopper in a real containment environment.

Sources and citation

Material revision history

  1. Sep 19, 2026 · Published version · first publication · revision 15
  2. Sep 19, 2026 · Published version · material revision · revision 20Public fields changed: attack_path, implications, mitigations, narrative, unknowns
  3. Sep 25, 2026 · Published version · material revision · revision 36Public fields changed: assessment_profile