Your Brain Has an Attack Surface

LessWrong published a finding from AI safety research that challenges how we think about detecting covert communication: signals don't escape detection by being more complex, but by moving through space. In experiments using neural networks, researchers sent hidden messages that evaded a monitor even though the messages remained completely readable to a retraining detector. The trick: signal clusters shifted by an average of two point five units in latent space while keeping their internal structure intact — like constellations gliding across the sky. A detector trained on the old positions failed, dropping from one hundred percent accuracy to about thirty percent. The researcher is extending this to brain-computer interfaces, asking whether this geometric evasion works regardless of whether the substrate is a neural network or brain signals, though real-world validation is still pending.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton