arXiv subjects

Taekyoung Kwon

Publications and source records attributed to Taekyoung Kwon.

8 recordsLinked to original sources

arXiv2026

MEMORY Wins All: Indirect Bias Injection Attacks via Social Media Feeds

Personal AI agents routinely consume external content while performing tasks such as web browsing, email processing, and SNS feed summarization, and they retain selected information or execution results in persistent memory for later use. We show that this ordinary ingestion of external content opens an indirect path for manipulating subsequent agent behavior. Based on this observation, we present IBIA, an Indirect Bias Injection Attack that plants an adversary-aligned stance on a specific topic into a victim agent's memory through external content, without direct access to the agent, its memory, or future user queries. For this, IBIA combines three mechanisms: comment cloaking, which keeps the crafted content consistent with the surrounding discussion, comment watermarking, which enables lightweight identification during curation, and category anchoring, which makes the retained stance salient under later related requests. We evaluate IBIA on BiasBench, a benchmark of 6,000 adversary-crafted social comments and 120 email instances. The watermark-based curation identifies 95.9% of the injected comments. Under the OpenClaw setting, IBIA achieves adversary-aligned response rates (AARs) of 91.2% on average across four downstream tasks, including 86.6% on the frontier GPT-5.5. We further propose a memory boundary defense that detects the injected bias and reduces AARs to 80.6%.

cs.AI↗

arXiv2026

DKVE: Decentralized Key Validation for End-to-End Encrypted Messaging

End-to-end encrypted messaging systems depend on authentic public key distribution to prevent man-in-the-middle (MitM) attacks. Current solutions present a stark trade-off: out-of-band (OOB) verification provides strong security but lacks scalability for large contact lists, while key transparency (KT) systems enable automated verification at high storage costs and operational complexity. We propose DKVE, a protocol that validates public keys through privacy-preserving cross-validation within users' social graphs. When obtaining a contact's public key from a key server, clients query mutual contacts to verify they hold the same key, combining Oblivious Pseudorandom Functions (OPRF) and Oblivious Key-Value Stores (OKVS) to preserve privacy of both queries and contact lists. DKVE employs a Sequential Probability Ratio Test (SPRT) to aggregate responses and detect server misbehavior with user-configurable error bounds. We evaluate DKVE through simulations on real social network datasets, demonstrating DKVE can detect MitM attacks with exceeding 97% for strong-to-moderate-tie networks. The remaining 3% of cases require validation through alternative methods such as KT and OOB verification. Our proof-of-concept implementation confirms feasibility for background operation on commodity hardware, in terms of the latency and bandwidth. As DKVE can reduce the frequency of KT queries by two orders of magnitude, it enables fundamental architectural shifts: KT directories can migrate from fast but space-inefficient Merkle trees to space-efficient data structures like RSA accumulators. While DKVE cannot replace existing methods entirely -- suffering from bootstrapping problems and degraded performance on weak-tie networks -- it provides a practical complementary key validation mechanism, making secure messaging more deployable for billion-user systems.

cs.CR↗

arXiv2026

TESLA-for-5G: Broadcast Authentication for 5G Networks Using TESLA

5G base stations broadcast unauthenticated system information (SI) that every user equipment (UE) reads during cell selection. This enables attackers to broadcast forged SI from a fake base station (FBS), deceiving UEs into camping on it. Prior approaches require UEs to authenticate System Information Block 1 (SIB1) using digital signatures. This necessitates computation-heavy verification for every SIB1 reception, imposing a significant burden on resource-constrained UEs. We propose TESLA-for-5G (TF5), a broadcast authentication protocol for 5G SIB1 that combines TESLA with GG09 Schnorr-like identity-based signatures (IBS). In the steady state, TF5 enables UEs to authenticate each SIB1 message using a symmetric MAC and delayed key disclosure, eliminating the need for per-message digital signatures. Initial trust is bootstrapped during cell entry using a lightweight GG09 IBS over the TESLA parameters, avoiding certificate distribution overhead. We formally verify TF5 in Tamarin under a Dolev-Yao adversary and demonstrate its favorable computation, communication, and storage costs through both an implementation on the OpenAirInterface 5G stack and trace-driven analysis.

cs.CR↗

arXiv2025

Multi-View Slot Attention Using Paraphrased Texts for Face Anti-Spoofing

Recent face anti-spoofing (FAS) methods have shown remarkable cross-domain performance by employing vision-language models like CLIP. However, existing CLIP-based FAS models do not fully exploit CLIP's patch embedding tokens, failing to detect critical spoofing clues. Moreover, these models rely on a single text prompt per class (e.g., 'live' or 'fake'), which limits generalization. To address these issues, we propose MVP-FAS, a novel framework incorporating two key modules: Multi-View Slot attention (MVS) and Multi-Text Patch Alignment (MTPA). Both modules utilize multiple paraphrased texts to generate generalized features and reduce dependence on domain-specific text. MVS extracts local detailed spatial features and global context from patch embeddings by leveraging diverse texts with multiple perspectives. MTPA aligns patches with multiple text representations to improve semantic robustness. Extensive experiments demonstrate that MVP-FAS achieves superior generalization performance, outperforming previous state-of-the-art methods on cross-domain datasets. Code: https://github.com/Elune001/MVP-FAS.

cs.CV↗

arXiv2024

Adversarial Feature Alignment: Balancing Robustness and Accuracy in Deep Learning via Adversarial Training

Deep learning models continue to advance in accuracy, yet they remain vulnerable to adversarial attacks, which often lead to the misclassification of adversarial examples. Adversarial training is used to mitigate this problem by increasing robustness against these attacks. However, this approach typically reduces a model's standard accuracy on clean, non-adversarial samples. The necessity for deep learning models to balance both robustness and accuracy for security is obvious, but achieving this balance remains challenging, and the underlying reasons are yet to be clarified. This paper proposes a novel adversarial training method called Adversarial Feature Alignment (AFA), to address these problems. Our research unveils an intriguing insight: misalignment within the feature space often leads to misclassification, regardless of whether the samples are benign or adversarial. AFA mitigates this risk by employing a novel optimization algorithm based on contrastive learning to alleviate potential feature misalignment. Through our evaluations, we demonstrate the superior performance of AFA. The baseline AFA delivers higher robust accuracy than previous adversarial contrastive learning methods while minimizing the drop in clean accuracy to 1.86% and 8.91% on CIFAR10 and CIFAR100, respectively, in comparison to cross-entropy. We also show that joint optimization of AFA and TRADES, accompanied by data augmentation using a recent diffusion model, achieves state-of-the-art accuracy and robustness.

cs.CV↗

arXiv2024

Amplifying Training Data Exposure through Fine-Tuning with Pseudo-Labeled Memberships

Neural language models (LMs) are vulnerable to training data extraction attacks due to data memorization. This paper introduces a novel attack scenario wherein an attacker adversarially fine-tunes pre-trained LMs to amplify the exposure of the original training data. This strategy differs from prior studies by aiming to intensify the LM's retention of its pre-training dataset. To achieve this, the attacker needs to collect generated texts that are closely aligned with the pre-training data. However, without knowledge of the actual dataset, quantifying the amount of pre-training data within generated texts is challenging. To address this, we propose the use of pseudo-labels for these generated texts, leveraging membership approximations indicated by machine-generated probabilities from the target LM. We subsequently fine-tune the LM to favor generations with higher likelihoods of originating from the pre-training data, based on their membership probabilities. Our empirical findings indicate a remarkable outcome: LMs with over 1B parameters exhibit a four to eight-fold increase in training data exposure. We discuss potential mitigations and suggest future research directions.

cs.CL↗

arXiv2024

CovRL: Fuzzing JavaScript Engines with Coverage-Guided Reinforcement Learning for LLM-based Mutation

Fuzzing is an effective bug-finding technique but it struggles with complex systems like JavaScript engines that demand precise grammatical input. Recently, researchers have adopted language models for context-aware mutation in fuzzing to address this problem. However, existing techniques are limited in utilizing coverage guidance for fuzzing, which is rather performed in a black-box manner. This paper presents a novel technique called CovRL (Coverage-guided Reinforcement Learning) that combines Large Language Models (LLMs) with reinforcement learning from coverage feedback. Our fuzzer, CovRL-Fuzz, integrates coverage feedback directly into the LLM by leveraging the Term Frequency-Inverse Document Frequency (TF-IDF) method to construct a weighted coverage map. This map is key in calculating the fuzzing reward, which is then applied to the LLM-based mutator through reinforcement learning. CovRL-Fuzz, through this approach, enables the generation of test cases that are more likely to discover new coverage areas, thus improving vulnerability detection while minimizing syntax and semantic errors, all without needing extra post-processing. Our evaluation results indicate that CovRL-Fuzz outperforms the state-of-the-art fuzzers in terms of code coverage and bug-finding capabilities: CovRL-Fuzz identified 48 real-world security-related bugs in the latest JavaScript engines, including 39 previously unknown vulnerabilities and 11 CVEs.

cs.CR↗

arXiv2011

Decryptable to Your Eyes: Visualization of Security Protocols at the User Interface

The design of authentication protocols, for online banking services in particular and any service that is of sensitive nature in general, is quite challenging. Indeed, enforcing security guarantees has overhead thus imposing additional computation and design considerations that do not always meet usability and user requirements. On the other hand, relaxing assumptions and rigorous security design to improve the user experience can lead to security breaches that can harm the users' trust in the system. In this paper, we demonstrate how careful visualization design can enhance not only the security but also the usability of the authentication process. To that end, we propose a family of visualized authentication protocols, a visualized transaction verification, and a "decryptable to your eyes only" protocol. Through rigorous analysis, we verify that our protocols are immune to many of the challenging authentication attacks applicable in the literature. Furthermore, using an extensive case study on a prototype of our protocols, we highlight the potential of our approach for real-world deployment: we were able to achieve a high level of usability while satisfying stringent security requirements.

cs.CR↗