Vibration Side-Channel Bypasses 3D Printer Acoustic Defense at 61%

Active Motor Noise Cancellation (AMNC) fully neutralizes acoustic side-channel attacks on Bambu Lab FDM printers, dropping classification accuracy to the 8.33% random baseline. The vibration channel, which AMNC ignores, still leaks a surprising amount of information about the printed object.

Acoustic Channel Dead, But Vibration Carries the Signal

Bambu Lab ships AMNC as a hardware countermeasure against acoustic side-channel attacks that steal intellectual property from 3D printers. A new empirical evaluation from an anonymous team (arXiv:2606.13952) confirms AMNC works perfectly on the acoustic front: their classifier could do no better than guessing at random across 12 object classes. That is the good news.

The bad news: the vibration channel is completely unprotected. Using only simple summary statistics from vibration recordings, the authors achieved around 31% pooled accuracy and 36-47% within a single printer. That is coarse, amplitude-driven leakage - enough to tell that something is being printed, but not what.

Temporal Model Squeezes Out Geometry Details

Vibration waveforms carry shape information when you look at how they evolve over the full print. A temporal model that ingests the ordered sequence of vibration frames pushes accuracy to roughly 61%. An order-shuffling control drops back to about 33%, proving the sequential structure of the print is the key signal, not just raw amplitude.

Critically, the leak is device-specific. A classifier trained on one Bambu Lab printer transfers near chance to a second unit. This means vibration fingerprints are tied to the mechanical idiosyncrasies of each machine, not just the object geometry.

What This Means for IP Protection

AMNC is an acoustic-only band-aid, not a comprehensive side-channel shield. The authors note that reconstruction-grade attacks would require the magnetic or power channels AMNC also leaves untouched. For now, vibration gives an adversary a partial, geometry-correlated view of the print but falls well short of full geometric reconstruction.

The team has released all code and the public dataset of synchronized acoustic and vibration recordings from two AMNC-equipped Bambu Lab printers. That dataset now gives the security community a concrete baseline to see just how much work is left before 3D printer IP theft is truly blocked.

Source: Side-Channel Attacks Bypass Protection in 3D Printers
Domain: arxiv.org