Brain-Spoofing & Signal Injection
Attacks on the sensing front-end itself, in which an adversary introduces false neural-looking signals so the system decodes activity that the brain did not produce. Vectors include electromagnetic interference coupled into leads, acoustic or ultrasonic injection into MEMS sensors, and tampering upstream of the decoder in software. The goal is a false read (spoofing) or, against a closed-loop stimulator, a triggered action.
Spoofing is most consequential in closed-loop systems where a decoded state drives stimulation, because an injected signal can propagate to a physical intervention. Defenses draw on sensor-security practice: physiological-plausibility and liveness checks, redundant and heterogeneous sensing, shielding, and cryptographic integrity on the sensor-to-decoder path so injected samples can be rejected.
Distinct from adversarial examples, which perturb a genuine signal; injection fabricates signal at the physical layer and is generally harder to mount but bypasses model-level defenses.