DiskFiltration encoded secrets in hard-drive actuator noise

Reconstruction

DiskFiltration begins after malware is already running on an air-gapped machine that contains a mechanical hard-disk drive. Instead of using speakers, the malware deliberately issues disk operations that move the HDD actuator arm in controlled patterns. Those mechanical movements create acoustic emissions at frequencies that can be selected and modulated by software.

A nearby microphone-equipped device records the sound and decodes the modulation. The important physical fact is that a hard drive is not only a storage device: its actuator is an electromechanical component whose motion produces sound. If software can control the motion precisely enough, an unintended mechanical side effect becomes a low-bandwidth transmitter. The paper reports 180 bits per minute and distances up to two meters in its tested setups.

The result is bounded. It requires prior compromise, a controllable mechanical HDD, suitable acoustic conditions and a receiver within range. It does not apply to SSD-only systems, and it does not demonstrate remote compromise or bulk network-like throughput.

Mechanism & boundary

1. 01

   Gain code execution on the isolated host

   DiskFiltration assumes malware is already present and can generate storage operations; it is not an initial-access technique.

   Boundary: air-gapped host / attacker-controlled code

2. 02

   Translate data into disk-access patterns

   The malware encodes secret bits into a sequence of HDD operations chosen to control actuator movement.

   Boundary: secret data / storage workload

3. 03

   Move the actuator arm deliberately

   Seek activity drives the mechanical actuator, producing controllable acoustic emissions without speakers or audio hardware.

   Boundary: software-controlled I/O / electromechanical motion

4. 04

   Capture the acoustic signal nearby

   A smartphone, smartwatch, laptop or other microphone-equipped receiver records the emitted pattern within the practical range.

   Boundary: protected machine / surrounding acoustic environment

5. 05

   Demodulate the secret

   Receiver software identifies the encoded acoustic pattern and reconstructs the transmitted data.

   Boundary: microphone recording / recovered data

Claims & evidence

Implications

The demonstrated impact is low-bandwidth exfiltration of compact secrets from an already-compromised HDD-based system. The broader lesson is that removing speakers does not create an 'audio gap' when other mechanical components remain software-controllable. For modern environments the applicability question is straightforward: SSD-only systems remove this specific actuator channel, while systems retaining mechanical disks may still inherit the physical prerequisite.

Controls & mitigations

• Prefer SSD-only storage for systems where the threat model justifies eliminating this mechanical actuator channel.
• Keep untrusted microphone-equipped devices outside the protected hardware zone so a practical receiver is not available.
• Use acoustic monitoring only where the environment and risk justify it, focusing on repeatable patterned emissions rather than ordinary disk noise.
• Constrain untrusted code execution and suspicious storage workloads because the channel requires software control of HDD activity.

What remains unknown

• The cited work does not establish practicality on systems without controllable mechanical HDDs.
• Range and decoding reliability depend on drive model, chassis acoustics, ambient noise and receiver placement.
• The experiment demonstrates exfiltration after compromise; it does not establish a method for initially infecting an air-gapped host.
• No cited evidence shows an AI system autonomously selecting or operating this channel in deployment.

Connected research