1 paper across 1 session
We use mechanistic interpretability to reverse engineer how neural networks break protected cryptographic implementations via side-channel analysis.
