Indistinguishability is a fundamental principle of cryptographic security, crucial for securing data transmitted between Internet of Things (IoT) devices. This principle ensures that an attacker cannot distinguish between the encrypted data, also known as ciphertext, and random data or the ciphertexts of the two messages encrypted with the same key. This research investigates the ability of machine learning (ML) in assessing indistinguishability property in encryption systems, with a focus on lightweight ciphers. As our first case study, we consider the SPECK32/64 and SIMON32/64 lightweight block ciphers, designed for IoT devices operating under significant energy constraints. In this research, we introduce MIND-Crypt1, a novel MLbased framework designed to assess the cryptographic indistinguishability of lightweight block ciphers, specifically the SPECK32/64 and SIMON32/64 encryption algorithm in CBC mode (Cipher Block Chaining), under Known Plaintext Attacks (KPA). Our approach involves training ML models using ciphertexts from two plaintext messages encrypted with same key to determine whether ML algorithms can identify meaningful cryptographic patterns or leakage. Our experiments show that modern ML techniques consistently achieve accuracy equivalent to random guessing, indicating that no statistically exploitable patterns exists in the ciphertexts generated by considered lightweight block ciphers. Furthermore, we demonstrate that in ML algorithms with all the possible combinations of the ciphertexts for given plaintext messages reflects memorization rather than generalization to unseen ciphertexts. Collectively, these findings suggest that existing block ciphers have secure cryptographic designs against ML-based indistinguishability assessments, reinforcing their security even under round-reduced conditions.1We refer to our attack framework as MIND-Crypt which stands for “Machine learning based framework for assessing INDistinguishability of Cryptographic Algorithms.”

While advertising has become commonplace in today's online interactions, there is a notable dearth of research investigating the extent to which browser fingerprinting is harnessed for user tracking and targeted advertising. Prior studies only measured whether fingerprinting-related scripts are being run on the websites but that in itself does not necessarily mean that fingerprinting is being used for the privacy-invasive purpose of online tracking because fingerprinting might be deployed for the defensive purposes of bot/fraud detection and user authentication. It is imperative to address the mounting concerns regarding the utilization of browser fingerprinting in the realm of online advertising.

Website fingerprinting can reveal which sensitive website a user visits over encrypted network traffic. Obfuscating encrypted traffic, e.g., adding dummy packets, is considered as a primary approach to defend against website fingerprinting. How-ever, existing defenses relying on traffic obfuscation are either ineffective or introduce significant overheads. As recent website fingerprinting attacks heavily rely on deep neural networks to achieve high accuracy, producing adversarial examples could be utilized as a new way to obfuscate encrypted traffic. Unfortunately, existing adversarial example algorithms are designed for images and do not consider unique challenges for network traffic.In this paper, we design a new method, named AdvTraffic, which can customize perturbations produced by any existing adversarial example algorithm on images and derive adversarial examples over encrypted traffic. Our experimental results show that the integration of AdvTraffic, particularly with Generative Adversarial Networks, can effectively mitigate the accuracy of website fingerprinting from 95.0% to 10.2%, even if an attacker retrains a classifier with defended traffic. Compared to other defenses, our method outperforms most of them in mitigating attack accuracy and offers the lowest bandwidth overhead.

Website fingerprinting attacks can infer which website a user visits over encrypted network traffic. Recent studies can achieve high accuracy (e.g., 98%) by leveraging deep neural networks. However, current attacks rely on enormous encrypted traffic data, which are time-consuming to collect. Moreover, large-scale encrypted traffic data also need to be recollected frequently to adjust the changes in the website content. In other words, the bootstrap time for carrying out website fingerprinting is not practical. In this paper, we propose a new method, named Adaptive Fingerprinting, which can derive high attack accuracy over few encrypted traffic by leveraging adversarial domain adaption. With our method, an attacker only needs to collect few traffic rather than large-scale datasets, which makes website fingerprinting more practical in the real world. Our extensive experimental results over multiple datasets show that our method can achieve 89% accuracy over few encrypted traffic in the closed-world setting and 99% precision and 99% recall in the open-world setting. Compared to a recent study (named Triplet Fingerprinting), our method is much more efficient in pre-training time and is more scalable. Moreover, the attack performance of our method can outperform Triplet Fingerprinting in both the closed-world evaluation and open-world evaluation.