Fansmitter addresses the same 'speakerless' containment assumption from a different component. After compromise, software changes the rotation speed of CPU or chassis fans. Different speeds produce different acoustic signatures, allowing the attacker to encode bits in ordinary fan noise and recover them with a nearby microphone. The mechanism works because…
Inspect the ClaimsCase · DiggingBeagle record
Fansmitter leaked data through computer fan noise without speakers
Researchers showed that software-controlled fan speed can modulate acoustic emissions and transmit data from an audio-less air-gapped computer to a nearby microphone.
Scope
Laboratory acoustic covert-channel demonstration requiring compromised code and a nearby microphone receiver. Not an observed AI behavior.
Assessment profile
Assessment method- Setting
- research demonstration
- Exploitation
- demonstrated
- Evidence
- Grade B
Basis and provenance
Recorded fan-control and smartphone-reception measurements describe a laboratory demonstration, not deployed AI behavior. 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.
Evidence basis: CLM-FANSMITTER-RANGE · CLM-FANSMITTER-MECHANISM · Fansmitter: Acoustic Data Exfiltration from (Speakerless) Air-Gapped Computers
At a glance
Fansmitter demonstrates that eliminating audio peripherals does not eliminate every acoustic output. The risk is most relevant where a compromised system can control cooling hardware and an attacker can position a microphone nearby. It also generalizes defensively: support components such as fans should be evaluated as software-controlled physical…
Read the implicationsReal-world reliability depends on fan model, speed-control granularity, ambient noise, room acoustics and receiver position. The published maximum range and rate are experimental results rather than guarantees for arbitrary systems. No cited evidence shows autonomous AI use of Fansmitter in a deployed containment environment.
Limits and uncertaintyFull account
Fansmitter addresses the same 'speakerless' containment assumption from a different component. After compromise, software changes the rotation speed of CPU or chassis fans. Different speeds produce different acoustic signatures, allowing the attacker to encode bits in ordinary fan noise and recover them with a nearby microphone.
The mechanism works because cooling fans are physical actuators under at least partial software control. Their rotational speed changes the spectrum and amplitude of the sound they generate. A component installed only for thermal management can therefore become a crude acoustic transmitter even when the computer has no speakers or audio hardware. The reported prototype reached a smartphone at 0-8 meters and up to 900 bits per hour.
This is a constrained laboratory covert channel. Practical success depends on fan hardware, controllability, environmental noise and receiver placement; it is not evidence that arbitrary fan-equipped systems are remotely readable.
Mechanism and trust boundary
- 01
Run malware on the protected device
The sender must already execute attacker-controlled code and have enough influence over fan behavior.
Boundary: protected host / attacker-controlled process
- 02
Encode data as fan-speed states
Secret bits are mapped to changes in fan rotation speed or acoustic frequency.
Boundary: secret data / cooling-control behavior
- 03
Generate modulated fan noise
The mechanical fan turns control changes into audible acoustic emissions even though no speaker is present.
Boundary: fan actuator / room acoustics
- 04
Record with a nearby microphone
A phone or other receiver captures the fan-noise pattern from within the tested physical range.
Boundary: protected room / receiver sensor
- 05
Recover the bitstream
Receiver software analyzes the acoustic signature and reconstructs the encoded data.
Boundary: acoustic recording / recovered secret
Claims & evidence
CLM-FANSMITTER-RANGEThe preprint reports successful transmission to a smartphone receiver over 0-8 meters at up to 900 bits per hour in the tested setup.supported
Basis: reported finding
- supportsFansmitter: Acoustic Data Exfiltration from (Speakerless) Air-Gapped Computersrole not specified
Abstract, reported range and rate
CLM-FANSMITTER-MECHANISMFansmitter demonstrated acoustic data exfiltration by controlling CPU/chassis fan speed even when the computer had no speakers or audio hardware.supported
Basis: reported finding
- supportsFansmitter: Acoustic Data Exfiltration from (Speakerless) Air-Gapped Computersrole not specified
Abstract, mechanism description
Implications
Fansmitter demonstrates that eliminating audio peripherals does not eliminate every acoustic output. The risk is most relevant where a compromised system can control cooling hardware and an attacker can position a microphone nearby. It also generalizes defensively: support components such as fans should be evaluated as software-controlled physical actuators, not only as maintenance hardware.
Controls and mitigations
- Remove untrusted microphone-equipped devices from high-assurance rooms and enforce physical separation from fan-equipped protected systems.
- Restrict or mediate software fan-control interfaces where operationally feasible.
- Monitor anomalous fan-speed patterns or acoustic signatures only where normal thermal-management variability can be distinguished reliably.
Unknowns and contradictions
- Real-world reliability depends on fan model, speed-control granularity, ambient noise, room acoustics and receiver position.
- The published maximum range and rate are experimental results rather than guarantees for arbitrary systems.
- No cited evidence shows autonomous AI use of Fansmitter in a deployed containment environment.
Sources and citation
Material revision history
- Sep 19, 2026 · Published version · first publication · revision 15
- Sep 19, 2026 · Published version · material revision · revision 20Public fields changed: attack_path, implications, mitigations, narrative, unknowns
- Sep 25, 2026 · Published version · material revision · revision 36Public fields changed: assessment_profile