| Popis: |
In 2020, Citizen Lab in University of Toronto found that the popular video conferencing application, Zoom, was not only ignoring the use of AES-256, but was instead using AES-128 in the Electronic Code Book (ECB) mode. ECB has never been a secure encryption strategy as it preserves the patterns present in the original input. This paper shows a simple deep learning based attack on images encrypted using four different encryption schemes (AES-128, GIFT-64, PICO and PRESENT) under the ECB mode of operation. We build a deep learning model that learns to classify encrypted images with 100% accuracy. We also explore the reasons behind why small size images might not look intelligible to the human eye but are still easily classifiable using a neural network. We also express how this attack can be scaled and further used to exploit encrypted data. |
