ODINI exfiltrated through CPU-generated magnetic fields despite Faraday shielding

Reconstruction

ODINI examines a containment layer stronger than an ordinary air gap: a computer placed behind metal shielding intended to suppress electromagnetic leakage. Malware varies the load on CPU cores in a controlled pattern. Changes in current through the processor and its power-delivery path alter the low-frequency magnetic field around the machine, allowing data to be encoded in that field.

The physical distinction matters. A conventional Faraday enclosure is effective against many electric-field and higher-frequency electromagnetic emissions, but low-frequency magnetic fields can penetrate metal shielding differently. The researchers used a nearby magnetic receiver to detect the modulated field and reported that their prototype did not require special privileges and could operate from an isolated virtual machine under their tested conditions.

ODINI does not prove that every Faraday-caged system leaks usable data. Receiver placement, field strength, shielding design, workload noise and hardware topology remain decisive. It demonstrates that shielding has to be designed for the relevant physical field rather than treated as a universal communications barrier.

Mechanism & boundary

1. 01

   Run code inside the protected machine

   The channel assumes attacker-controlled code is already executing, including in the researchers' tested VM scenario.

   Boundary: shielded host / attacker-controlled workload

2. 02

   Modulate CPU-core load

   The malware varies processor utilization according to the data it wants to transmit.

   Boundary: secret data / electrical workload

3. 03

   Generate a low-frequency magnetic pattern

   Changing processor current produces corresponding changes in the surrounding magnetic field.

   Boundary: electrical activity / magnetic field

4. 04

   Cross the shielding boundary

   The low-frequency magnetic component can propagate through shielding that is effective against other electromagnetic emissions.

   Boundary: shielded enclosure / external physical space

5. 05

   Measure with a nearby magnetic sensor

   A prepared receiver observes the field changes and decodes the transmitted symbols.

   Boundary: magnetic field / recovered data

Timeline

1. Feb 8, 2018

   ODINI preprint released

   report

   The researchers published the magnetic covert-channel design and prototype results.

2. Aug 29, 2019

   Peer-reviewed publication metadata recorded

   report

   The corresponding IEEE Transactions publication record was added for the peer-reviewed version.

Claims & evidence

Implications

ODINI's demonstrated impact is a local magnetic exfiltration path after compromise. Its broader implication is architectural: 'inside a Faraday cage' is not a complete statement of physical isolation unless the enclosure is characterized for the frequencies and field types an attacker can induce. The result should not be generalized to universal success through arbitrary shielded facilities.

Controls & mitigations

• Use physical exclusion zones that prevent untrusted magnetic sensors from being positioned close enough to protected equipment.
• Characterize shielding against relevant low-frequency magnetic fields rather than assuming high-frequency RF attenuation is sufficient.
• Monitor suspicious patterned CPU workloads where the assurance level justifies side-channel detection.

What remains unknown

• The cited work does not establish universal success across Faraday cages, rack layouts, processor generations or facility-scale shielding.
• Practical range and signal quality depend on hardware, receiver sensitivity, placement and environmental magnetic noise.
• No cited evidence establishes autonomous AI use of ODINI.

Connected research