MLPs Can Hide Memorized Secrets From Both Human and AI Extraction
Training multilayer perceptrons with hard‑negative examples and depth greater than one lets them memorize binary strings that remain undetectable by state‑of‑the‑art input‑optimization attacks and even transformer‑based weight‑readers, despite near‑perfect memorization. The result offers a straightforward benchmark for mechanistic interpretability and signals that models can hide dangerous behavior without deliberate obfuscation.