HeeSeok Kim

The widespread deployment of AI applications on wearable and IoT devices has raised security concerns, including model stealing and personal information leakage. Model stealing attacks rely on issuing queries to the target model and analyzing the corresponding outputs. Side-channel analysis, traditionally used to attack cryptographic algorithms, can now facilitate the performance of model stealing attacks on neural networks (NN), by recovering their architectures and parameters. This exposure to model stealing presents risks such as intellectual property theft and adversarial attacks. Masking, a widely used countermeasure for side-channel attacks, has recently been adapted to NN to protect secret model information. However, existing masked NNs incur significant computational overhead or lead to accuracy degradation. In this work, we focus on reducing the main degradation in masking computational costs of non-linear operations specifically activation and argmax. As both operations hinge on sign bit extraction, we achieved an efficient realization using LuTs. Furthermore, we systematically identify and remove flawed masking gadgets that cause accuracy degradation and subsequently apply our gadgets to binarized neural network (BNN). As a result, our proposed BNN implementation securely maintains model accuracy while achieving a 42% performance improvement compared to state-of-the-art software-based masking methods [AWDF21].

Despite advancements in password policy enforcement and user awareness, individuals continue to follow predictable patterns when updating or creating new passwords across different sites-often making minor changes such as appending digits, capitalizing letters, or adding special characters. To model and exploit such behavioral regularities, we present PentaPassBreaker, a novel sequence-to-sequence generative model for password prediction. As a foundation for learning real-world update patterns, we introduce a large-scale lexical similarity filtering of leaked credentials, extracting 307 million high-similarity password pairs through noise cleaning and similarity-based clustering. Our analysis reveals that users reuse an average of 3.2 closely related passwords per account, highlighting the prevalence of fine-grained reuse behavior. Experimental results show that PentaPassBreaker achieves an average top-5 accuracy of 22.1%, with simple transformation paths exceeding 35% in top-25 accuracy. Qualitative analysis further reveals that even incorrect guesses often resemble plausible user updates, demonstrating strong behavioral fidelity. Moreover, PPB consistently outperforms strong rule-based baselines such as John the Ripper (JTR), particularly on structural and multi-edit transformation paths under strict top-$k$ budgets. These findings highlight the overlooked risks associated with incremental password reuse and emphasize the need for stronger user awareness and authentication safeguards.

Hardware wallets, specialized devices designed to securely manage users’ credentials, play a crucial role in securing cryptocurrencies, ensuring credentials remain under user control without reliance on third-party entities. However, despite extensive research on Side-Channel Analysis (SCA) attacks, studies specifically addressing their implications for hardware wallets remain relatively limited. While previous work has demonstrated various SCA attacks on hardware wallets, most of these attacks require sophisticated environmental controls or detailed knowledge of target device. In addition, some attacks assume unrealistic scenarios that require valid credentials to conduct the attacks. This paper introduces a novel SCA attack on hardware wallets to extract master seeds—a foundational component in the security of hardware wallets. Our proposed attack leverages power traces obtained during the processing of the Keyed-Hash Message Authentication Code (HMAC), or more precisely, the Secure Hash Algorithm 2 (SHA-2) inside the HMAC. Notably, our attack is non-invasive, ensuring the integrity of the target device, thereby making it difficult for the wallet owners to detect the attack. Furthermore, our attack can be conducted without a profiling phase, excluding the excessive capabilities required for the attack.