Papers

Decentralized prediction markets (DePMs) allow open participation in event-based wagering without fully relying on centralized intermediaries. We review the history of DePMs which date back to 2011 and includes hundreds of proposals. Perhaps surprising, modern DePMs like Polymarket deviate materially from earlier designs like Truthcoin and Augur v1. We use our review to present a modular workflow comprising eight stages: underlying infrastructure, market topic, share structure and pricing, market initialization, trading, market resolution, settlement, and archiving. For each module, we enumerate the design variants, analyzing trade-offs around decentralization, expressiveness, and manipulation resistance. We also identify open problems for researchers interested in this ecosystem.

Malicious software attacks are having an increasingly significant economic impact. Commercial malware detection software can be costly, and tools that attribute malware to the specific software vulnerabilities it exploits are largely lacking. Understanding the connection between malware and the vulnerabilities it targets is crucial for analyzing past threats and proactively defending against current ones. In this study, we propose an approach that leverages large language models (LLMs) to detect binary malware, specifically within JAR files, and uses LLM capabilities combined with retrieval-augmented generation (RAG) to identify Common Vulnerabilities and Exposures (CVEs) that malware may exploit. We developed a proof-of-concept tool, MalCVE, that integrates binary code decompilation, deobfuscation, LLM-based code summarization, semantic similarity search, and LLM-based CVE classification. We evaluated MalCVE using a benchmark dataset of 3,839 JAR executables. MalCVE achieved a mean malware-detection accuracy of 97%, at a fraction of the cost of commercial solutions. In particular, the results demonstrate that LLM-based code summarization enables highly accurate and explainable malware identification. MalCVE is also the first tool to associate CVEs with binary malware, achieving a recall@10 of 65%, which is comparable to studies that perform similar analyses on source code.

This paper presents a novel post-quantum cryptosystem based on high-memory masked convolutional codes. Unlike conventional code-based schemes that rely on block codes with fixed dimensions and limited error-correction capability, our construction offers both stronger cryptographic security and greater flexibility. It supports arbitrary plaintext lengths with linear-time decryption and uniform per-bit computational cost, enabling seamless scalability to long messages. Security is reinforced through a higher-rate injection of random errors than in block-code approaches, along with additional noise introduced via polynomial division, which substantially obfuscates the underlying code structure. Semi-invertible transformations generate dense, random-like generator matrices that conceal algebraic properties and resist known structural attacks. Consequently, the scheme achieves cryptanalytic security margins exceeding those of the classic McEliece system by factors greater than 2100. Finally, decryption at the recipient employs an array of parallel Viterbi decoders, enabling efficient hardware and software implementation and positioning the scheme as a strong candidate for deployment in practical quantum-resistant public-key cryptosystems.

Recent advancements in Reinforcement Learning with Verifiable Rewards (RLVR) have gained significant attention due to their objective and verifiable reward signals, demonstrating strong performance in reasoning and code generation tasks. However, the potential safety risks associated with RLVR remain underexplored. This paper presents HarmRLVR, the first systematic investigation into the alignment reversibility risk of RLVR. We show that safety alignment can be rapidly reversed using GRPO with merely 64 harmful prompts without responses, causing models to readily comply with harmful instructions. Across five models from Llama, Qwen, and DeepSeek, we empirically demonstrate that RLVR-based attacks elevate the average harmfulness score to 4.94 with an attack success rate of 96.01%, significantly outperforming harmful fine-tuning while preserving general capabilities. Our findings reveal that RLVR can be efficiently exploited for harmful alignment, posing serious threats to open-source model safety. Please see our code at https://github.com/lyxx2535/HarmRLVR.

Prompt-security claims for large language models (LLMs) are often compared as properties of individual attacks or defenses, although measured harmfulness depends on the target model, access setting, attack and defense configuration, prompt sample, and judger. We systematize 272 studies through three linked, multi-label taxonomies of jailbreak attacks, defenses, and model vulnerabilities. To make these dependencies explicit, we introduce an access-compatible evaluation protocol over configured $\langle M,A,D,S,J\rangle$ tuples and instantiate it in PromptSecurity, an open-source platform that records method settings, outputs, judgments, and query costs. We release two complementary datasets: JailbreakDB, with 445,752 jailbreak system-user prompt pairs and 1,094,122 benign prompts, and PromptSecurity-Eval, with 166,200 harmful-query and 25,800 benign-utility output records. We evaluate 11 target models, 20 attacks, nine defenses, and three automated judgers under a matched, single-turn protocol using 100 harmful queries and separate benign-utility tasks. Lower attack success can reflect failure to execute transformed requests rather than stronger safety; attack-category advantages can reverse when one method is excluded, and attack success varies with assistant choice. Defenses can reduce harmful outputs while degrading benign-task utility or causing backfire. Rejudging identical outputs also shows that measured attack success is sensitive to the behavior supplied to the judger, highlighting the need to report judger configurations alongside ASR. These findings motivate configuration-aware reporting rather than context-free method rankings. Resources: https://datasec-lab.github.io/PromptSecurityLeaderboard/ https://huggingface.co/datasets/youbin2014/JailbreakDB https://huggingface.co/datasets/youbin2014/PromptSecurity-Eval https://github.com/datasec-lab/PromptSecurity

FHE-SQL is a privacy-preserving database system that enables secure query processing on encrypted data using Fully Homomorphic Encryption (FHE), providing privacy guaranties where an untrusted server can execute encrypted queries without learning either the query contents or the underlying data. Unlike property-preserving encryption-based systems such as CryptDB, which rely on deterministic or order-preserving encryption and are vulnerable to frequency, order, and equality-pattern inference attacks, FHE-SQL performs computations entirely under encryption, eliminating these leakage channels. Compared to trusted-hardware approaches such as TrustedDB, which depend on a hardware security module and thus inherit its trust and side-channel limitations, our design achieves end-to-end cryptographic protection without requiring trusted execution environments. In contrast to high-performance FHE-based engines-Hermes, which target specialized workloads such as vector search, FHE-SQL supports general SQL query semantics with schema-aware, type-safe definitions suitable for relational data management. FHE-SQL mitigates the high cost of ciphertext space by using an indirection architecture that separates metadata in RocksDB from large ciphertexts in blob storage. It supports oblivious selection via homomorphic boolean masks, multi-tier caching, and garbage collection, with security proven under the Universal Composability framework.

Beyond its widespread application in signal and image processing, compressed sensing principles have been greatly applied to secure information transmission (often termed 'compressive security'). In this scenario, the measurement matrix $Q$ acts as a one time pad encryption key (in complex number domain) which can achieve perfect information-theoretic security together with other benefits such as reduced complexity and energy efficiency particularly useful in IoT. However, unless the matrix is changed for every message it is vulnerable towards known plain text attacks: only $n$ observations suffices to recover a key $Q$ with $n$ columns. In this paper, we invent and analyze a new method (termed 'Bilinear Compressive Security (BCS)') addressing these shortcomings: In addition to the linear encoding of the message $x$ with a matrix $Q$, the sender convolves the resulting vector with a randomly generated filter $h$. Assuming that $h$ and $x$ are sparse, the receiver can then recover $x$ without knowledge of $h$ from $y=h*Qx$ through blind deconvolution. We study a rather idealized known plaintext attack for recovering $Q$ from repeated observations of $y$'s for different, known $x_k$, with varying and unknown $h$ ,giving Eve a number of advantages not present in practice. Our main result for BCS states that under a weak symmetry condition on the filter $h$, recovering $Q$ will require extensive sampling from transmissions of $\Omega\left(\max\left(n,(n/s)^2\right)\right)$ messages $x_k$ if they are $s$-sparse. Remarkably, with $s=1$ it is impossible to recover the key. In this way, the scheme is much safer than standard compressed sensing even though our assumptions are much in favor towards a potential attacker.

Federated learning (FL) is a distributed machine learning paradigm that enables multiple clients to collaboratively train a shared model without disclosing their local data. To address privacy issues of gradient, several privacy-preserving machine-learning schemes based on multi-client functional encryption (MCFE) have been proposed. However, existing MCFE-based schemes cannot support client dropout or flexible threshold selection, which are essential for practical FL. In this paper, we design a flexible threshold multi-client functional encryption for inner product (FTMCFE-IP) scheme, where multiple clients generate ciphertexts independently without any interaction. In the encryption phase, clients are able to choose a threshold flexibly without reinitializing the system. The decryption can be performed correctly when the number of online clients satisfies the threshold. An authorized user are allowed to compute the inner product of the vectors associated with his/her functional key and the ciphertext, respectively, but cannot learning anything else. Especially, the presented scheme supports clients drop out. Furthermore, we provide the definition and security model of our FTMCFE-IP scheme,and propose a concrete construction. The security of the designed scheme is formally proven. Finally, we implement and evaluate our FTMCFE-IP scheme.

Large web-scale datasets have driven the rapid advancement of pre-trained language models (PLMs), but unauthorized data usage has raised serious copyright concerns. Existing dataset ownership verification (DOV) methods typically assume that watermarks remain stable during inference; however, this assumption often fails under natural noise and adversary-crafted perturbations. We propose the first certified dataset ownership verification method for PLMs under a gray-box setting (i.e., the defender can only query the suspicious model but is aware of its input representation module), based on dual-space smoothing (i.e., DSSmoothing). To address the challenges of text discreteness and semantic sensitivity, DSSmoothing introduces continuous perturbations in the embedding space to capture semantic robustness and applies controlled token reordering in the permutation space to capture sequential robustness. DSSmoothing consists of two stages: in the first stage, triggers are collaboratively embedded in both spaces to generate norm-constrained and robust watermarked datasets; in the second stage, randomized smoothing is applied in both spaces during verification to compute the watermark robustness (WR) of suspicious models and statistically compare it with the principal probability (PP) values of a set of benign models. Theoretically, DSSmoothing provides provable robustness guarantees for dataset ownership verification by ensuring that WR consistently exceeds PP under bounded dual-space perturbations. Extensive experiments on multiple representative web datasets demonstrate that DSSmoothing achieves stable and reliable verification performance and exhibits robustness against potential adaptive attacks. Our code is available at https://github.com/NcepuQiaoTing/DSSmoothing.

When users submit queries to Large Language Models (LLMs), their prompts can often contain sensitive data, forcing a difficult choice: Send the query to a powerful proprietary LLM providers to achieving state-of-the-art performance and risk data exposure, or relying on smaller, local models guarantees data privacy but often results in a degradation of task performance. Prior approaches have relied on static pipelines that use LLM rewriting, which shatters linguistic coherence and indiscriminately removes privacy-sensitive information, including task-critical content. We reformulate this challenge (Privacy-Conscious Delegation) as a sequential decision-making problem and introduce a novel reinforcement learning (RL) framework called Privacy-R1 to solve it. Our framework trains an agent to dynamically route text chunks, learning a policy that optimally balances the trade-off between privacy leakage and task performance. It implicitly distinguishes between replaceable Personally Identifiable Information (PII) (which it shields locally) and task-critical PII (which it strategically sends to the remote model for maximal utility). To validate our approach in complex scenarios, we also introduce a new medical dataset with high PII density. Our framework achieves a new state-of-the-art on the privacy-utility frontier, demonstrating the necessity of learned, adaptive policies for deploying LLMs in sensitive environments. Dataset can be found at: https://github.com/zackhuiiiii/Privacy-R1.

Identification of anomalous events within system logs constitutes a pivotal element within the frame- work of cybersecurity defense strategies. However, this process faces numerous challenges, including the management of substantial data volumes, the distribution of anomalies, and the precision of con- ventional methods. To address this issue, the present paper puts forward a proposal for an intelligent detection method for system logs based on Genera- tive Pre-trained Transformers (GPT). The efficacy of this approach is attributable to a combination of structured input design and a Focal Loss op- timization strategy, which collectively result in a substantial enhancement of the performance of log anomaly detection. The initial approach involves the conversion of raw logs into event ID sequences through the use of the Drain parser. Subsequently, the Focal Loss loss function is employed to address the issue of class imbalance. The experimental re- sults demonstrate that the optimized GPT-2 model significantly outperforms the unoptimized model in a range of key metrics, including precision, recall, and F1 score. In specific tasks, comparable or superior performance has been demonstrated to that of the GPT-3.5 API.

This study investigates the privacy risks associated with diffusion-based synthetic tabular data generation methods, focusing on their susceptibility to Membership Inference Attacks (MIAs). We examine two recent models, TabDDPM and TabSyn, by developing query-based MIAs based on the step-wise error comparison method. Our findings reveal that TabDDPM is more vulnerable to these attacks. TabSyn exhibits resilience against our attack models. Our work underscores the importance of evaluating the privacy implications of diffusion models and encourages further research into robust privacy-preserving mechanisms for synthetic data generation.

Social networks have become a crucial source of real-time information for individuals. The influence of social bots within these platforms has garnered considerable attention from researchers, leading to the development of numerous detection technologies. However, the vulnerability and robustness of these detection methods is still underexplored. Existing Graph Neural Network (GNN)-based methods cannot be directly applied due to the issues of limited control over social agents, the black-box nature of bot detectors, and the heterogeneity of bots. To address these challenges, this paper proposes the first adversarial multi-agent Reinforcement learning framework for social Bot control attacks (RoBCtrl) targeting GNN-based social bot detectors. Specifically, we use a diffusion model to generate high-fidelity bot accounts by reconstructing existing account data with minor modifications, thereby evading detection on social platforms. To the best of our knowledge, this is the first application of diffusion models to mimic the behavior of evolving social bots effectively. We then employ a Multi-Agent Reinforcement Learning (MARL) method to simulate bots adversarial behavior. We categorize social accounts based on their influence and budget. Different agents are then employed to control bot accounts across various categories, optimizing the attachment strategy through reinforcement learning. Additionally, a hierarchical state abstraction based on structural entropy is designed to accelerate the reinforcement learning. Extensive experiments on social bot detection datasets demonstrate that our framework can effectively undermine the performance of GNN-based detectors.

Advanced Persistent Threats (APTs) are stealthy cyberattacks that often evade detection in system-level audit logs. Provenance graphs model these logs as connected entities and events, revealing relationships that are missed by linear log representations. Existing systems apply anomaly detection to these graphs but often suffer from high false positive rates and coarse-grained alerts. Their reliance on node attributes like file paths or IPs leads to spurious correlations, reducing detection robustness and reliability. To fully understand an attack's progression and impact, security analysts need systems that can generate accurate, human-like narratives of the entire attack. To address these challenges, we introduce OCR-APT, a system for APT detection and reconstruction of human-like attack stories. OCR-APT uses Graph Neural Networks (GNNs) for subgraph anomaly detection, learning behavior patterns around nodes rather than fragile attributes such as file paths or IPs. This approach leads to a more robust anomaly detection. It then iterates over detected subgraphs using Large Language Models (LLMs) to reconstruct multi-stage attack stories. Each stage is validated before proceeding, reducing hallucinations and ensuring an interpretable final report. Our evaluations on the DARPA TC3, OpTC, and NODLINK datasets show that OCR-APT outperforms state-of-the-art systems in both detection accuracy and alert interpretability. Moreover, OCR-APT reconstructs human-like reports that comprehensively capture the attack story.

A core challenge for autonomous LLM agents in collaborative settings is balancing robust privacy understanding and preservation alongside task efficacy. Existing privacy benchmarks only focus on simplistic, single-turn interactions where private information can be trivially omitted without affecting task outcomes. In this paper, we introduce MAGPIE (Multi-AGent contextual PrIvacy Evaluation), a novel benchmark of 200 high-stakes tasks designed to evaluate privacy understanding and preservation in multi-agent collaborative, non-adversarial scenarios. MAGPIE integrates private information as essential for task resolution, forcing agents to balance effective collaboration with strategic information control. Our evaluation reveals that state-of-the-art agents, including GPT-5 and Gemini 2.5-Pro, exhibit significant privacy leakage, with Gemini 2.5-Pro leaking up to 50.7% and GPT-5 up to 35.1% of the sensitive information even when explicitly instructed not to. Moreover, these agents struggle to achieve consensus or task completion and often resort to undesirable behaviors such as manipulation and power-seeking (e.g., Gemini 2.5-Pro demonstrating manipulation in 38.2% of the cases). These findings underscore that current LLM agents lack robust privacy understanding and are not yet adequately aligned to simultaneously preserve privacy and maintain effective collaboration in complex environments.

Educating children and young people to be safe online is essential, especially as the metaverse, a next-generation internet blending immersive technologies, promises to reshape their interactions and amplify their experiences. While virtual reality offers fully immersive, highly interactive, and multi-sensory engagement, it also heightens cyber harm risks for young or vulnerable users. To address this, the CyberNinjas VR experience was developed to educate children aged 8 to 16 on safe metaverse behaviours, providing clear referral steps for harmful interactions. Understanding user engagement in metaverse gaming will aid the design of future VR environments which prioritize safety and inclusivity. This project analyses CyberNinjas to understand how game mechanics can foster cyber-safe behaviours.

Voice interfaces are increasingly used in high-stakes domains such as mobile banking, smart-home security, and hands-free healthcare. Meanwhile, modern generative models have made high-quality voice forgeries inexpensive and easy to create, eroding confidence in voice authentication alone. To strengthen protection against such attacks, we present a second authentication factor that combines acoustic evidence with the unique motion patterns of a speaker's lower face. By placing lightweight inertial sensors around the mouth to capture mouth opening and evolving lower-facial geometry, our system records a distinct motion signature with strong discriminative power across individuals. We built a prototype and recruited 43 participants to evaluate the system under four conditions: seated, walking on level ground, walking on stairs, and speaking with different language backgrounds (native vs. non-native English). Across all scenarios, our approach consistently achieved a median equal-error rate (EER) of 0.01 or lower, indicating that mouth-movement data remain robust under variations in gait, posture, and spoken language. We discuss specific use cases where this second line of defense could provide tangible security benefits to voice authentication systems.

File-encrypting ransomware increasingly employs intermittent encryption techniques, encrypting only parts of files to evade classical detection methods.This paper provides a systematic empirical characterization of byte-level statistics under intermittent encryption across common file types, establishing a baseline for how partial encryption reshapes data structure. Guided by these measurements, we model intermittent encryption as a convex mixture of ciphertext and cleartext and, via a classical KL-divergence bound, derive file-type-specific detectability limits for histogram-based detectors. Leveraging these insights, we evaluate convolutional neural network (CNN) detectors trained on realistic intermittent-encryption configurations from leading ransomware families. Our findings show that localized, chunk-level CNNs consistently outperform whole-file analysis, highlighting a practical, robust baseline for future detection systems.

With the exponential growth in mobile applications, protecting user privacy has become even more crucial. Android applications are often known for collecting, storing, and sharing sensitive user information such as contacts, location, camera, and microphone data often without the user's clear consent or awareness raising significant privacy risks and exposure. In the context of privacy assessment, dataflow analysis is particularly valuable for identifying data usage and potential leaks. Traditionally, this type of analysis has relied on formal methods, heuristics, and rule-based matching. However, these techniques are often complex to implement and prone to errors, such as taint explosion for large programs. Moreover, most existing Android dataflow analysis methods depend heavily on predefined list of sinks, limiting their flexibility and scalability. To address the limitations of these existing techniques, we propose AndroByte, an AI-driven privacy analysis tool that leverages LLM reasoning on bytecode summarization to dynamically generate accurate and explainable dataflow call graphs from static code analysis. AndroByte achieves a significant Fe̱ta}-Score of 89% in generating dynamic dataflow call graphs on the fly, outperforming the effectiveness of traditional tools like FlowDroid and Amandroid in leak detection without relying on predefined propagation rules or sink lists. Moreover, AndroByte's iterative bytecode summarization provides comprehensive and explainable insights into dataflow and leak detection, achieving high, quantifiable scores based on the G-Eval metric.

This paper investigates the algebraic and graphical structure of the ring $\mathbb{Z}_{sp}$, with a focus on its decomposition into finite fields, kernels, and special subsets. We establish classical isomorphisms between $\mathbb{F}_s$ and $p\mathbb{F}_s$, as well as $p\mathbb{F}_s^{\star}$ and $p\mathbb{F}_s^{+1,\star}$. We introduce the notion of arcs and rooted trees to describe the pre-periodic structure of $\mathbb{Z}_{sp}$, and prove that trees rooted at elements not divisible by $s$ or $p$ can be generated from the tree of unity via multiplication by cyclic arcs. Furthermore, we define and analyze the set $\mathbb{D}_{sp}$, consisting of elements that are neither multiples of $s$ or $p$ nor "off-by-one" elements, and show that its graph decomposes into cycles and pre-periodic trees. Finally, we demonstrate that every cycle in $\mathbb{Z}_{sp}$ contains inner cycles that are derived predictably from the cycles of the finite fields $p\mathbb{F}_s$ and $s\mathbb{F}_p$, and we discuss the cryptographic relevance of $\mathbb{D}_{sp}$, highlighting its potential for analyzing cyclic attacks and factorization methods.

As Large Language Models (LLMs) gain traction across critical domains, ensuring secure and trustworthy training processes has become a major concern. Backdoor attacks, where malicious actors inject hidden triggers into training data, are particularly insidious and difficult to detect. Existing post-training verification solutions like Proof-of-Learning are impractical for LLMs due to their requirement for full retraining, lack of robustness against stealthy manipulations, and inability to provide early detection during training. Early detection would significantly reduce computational costs. To address these limitations, we introduce Proof-of-Training Steps, a verification protocol that enables an independent auditor (Alice) to confirm that an LLM developer (Bob) has followed the declared training recipe, including data batches, architecture, and hyperparameters. By analyzing the sensitivity of the LLMs' language modeling head (LM-Head) to input perturbations, our method can expose subtle backdoor injections or deviations in training. Even with backdoor triggers in up to 10 percent of the training data, our protocol significantly reduces the attacker's ability to achieve a high attack success rate (ASR). Our method enables early detection of attacks at the injection step, with verification steps being 3x faster than training steps. Our results highlight the protocol's potential to enhance the accountability and security of LLM development, especially against insider threats.

The Synthetic Minority Over-sampling Technique (SMOTE) is one of the most widely used methods for addressing class imbalance and generating synthetic data. Despite its popularity, little attention has been paid to its privacy implications; yet, it is used in the wild in many privacy-sensitive applications. In this work, we conduct the first systematic study of privacy leakage in SMOTE: we begin by showing that prevailing evaluation practices, i.e., naive distinguishing and distance-to-closest-record metrics, completely fail to detect any leakage and that membership inference attacks (MIAs) can be instantiated with high accuracy. Then, by exploiting SMOTE's geometric properties, we build two novel attacks with very limited assumptions: DistinSMOTE, which perfectly distinguishes real from synthetic records in augmented datasets, and ReconSMOTE, which reconstructs real minority records from synthetic datasets with perfect precision and recall approaching one under realistic imbalance ratios. We also provide theoretical guarantees for both attacks. Experiments on eight standard imbalanced datasets confirm the practicality and effectiveness of these attacks. Overall, our work reveals that SMOTE is inherently non-private and disproportionately exposes minority records, highlighting the need to reconsider its use in privacy-sensitive applications and as a baseline for assessing the privacy of modern generative models.

Multimodal large language models (MLLMs) exhibit remarkable capabilities but remain susceptible to jailbreak attacks exploiting cross-modal vulnerabilities. In this work, we introduce a novel method that leverages sequential comic-style visual narratives to circumvent safety alignments in state-of-the-art MLLMs. Our method decomposes malicious queries into visually innocuous storytelling elements using an auxiliary LLM, generates corresponding image sequences through diffusion models, and exploits the models' reliance on narrative coherence to elicit harmful outputs. Extensive experiments on harmful textual queries from established safety benchmarks show that our approach achieves an average attack success rate of 83.5%, surpassing prior state-of-the-art by 46%. Compared with existing visual jailbreak methods, our sequential narrative strategy demonstrates superior effectiveness across diverse categories of harmful content. We further analyze attack patterns, uncover key vulnerability factors in multimodal safety mechanisms, and evaluate the limitations of current defense strategies against narrative-driven attacks, revealing significant gaps in existing protections.

Physical layer deception (PLD) combines physical layer security (PLS) with deception: the transmitter actively misleads the eavesdropper with falsified information. We model the transmitter-eavesdropper interaction as a Stackelberg game in which the transmitter commits to a resource allocation and encryption strategy, and each receiver best-responds by selecting among three decryption modes: Perception, Dropping, and Exclusion. Using semantic distortion as the metric, we derive closed-form switching surfaces that partition the parameter space into strategy regimes and identify conditions under which each regime dominates. The robust operating point, at the peak of the worst-case distortion envelope, is shown to be a Stackelberg equilibrium; iterative best-response dynamics oscillate around it with strictly lower time-averaged security. We evaluate the design under Nakagami-m fading with static and adaptive transmitter strategies, benchmarked against a classical PLS baseline. Numerical results validate the regime characterization and show 12-55% higher eavesdropper distortion than the erasure-only baseline across all fading conditions.

Large language models (LLMs) are increasingly vulnerable to multi-turn jailbreak attacks, where adversaries iteratively elicit harmful behaviors that bypass single-turn safety filters. Existing defenses predominantly rely on passive rejection, which either fails against adaptive attackers or overly restricts benign users. We propose a honeypot-based proactive guardrail system that transforms risk avoidance into risk utilization. Our framework fine-tunes a bait model to generate ambiguous, non-actionable but semantically relevant responses, which serve as lures to probe user intent. Combined with the protected LLM's safe reply, the system inserts proactive bait questions that gradually expose malicious intent through multi-turn interactions. We further introduce the Honeypot Utility Score (HUS), measuring both the attractiveness and feasibility of bait responses, and use a Defense Efficacy Rate (DER) for balancing safety and usability. Initial experiment on MHJ Datasets with recent attack method across GPT-4o show that our system significantly disrupts jailbreak success while preserving benign user experience.

Machine Learning (ML)-based malicious traffic detection is a promising security paradigm. It outperforms rule-based traditional detection by identifying various advanced attacks. However, the robustness of these ML models is largely unexplored, thereby allowing attackers to craft adversarial traffic examples that evade detection. Existing evasion attacks typically rely on overly restrictive conditions (e.g., encrypted protocols, Tor, or specialized setups), or require detailed prior knowledge of the target (e.g., training data and model parameters), which is impractical in realistic black-box scenarios. The feasibility of a hard-label black-box evasion attack (i.e., applicable across diverse tasks and protocols without internal target insights) thus remains an open challenge. To this end, we develop NetMasquerade, which leverages reinforcement learning (RL) to manipulate attack flows to mimic benign traffic and evade detection. Specifically, we establish a tailored pre-trained model called Traffic-BERT, utilizing a network-specialized tokenizer and an attention mechanism to extract diverse benign traffic patterns. Subsequently, we integrate Traffic-BERT into the RL framework, allowing NetMasquerade to effectively manipulate malicious packet sequences based on benign traffic patterns with minimal modifications. Experimental results demonstrate that NetMasquerade enables both brute-force and stealthy attacks to evade 6 existing detection methods under 80 attack scenarios, achieving over 96.65% attack success rate. Notably, it can evade the methods that are either empirically or certifiably robust against existing evasion attacks. Finally, NetMasquerade achieves low-latency adversarial traffic generation, demonstrating its practicality in real-world scenarios.

The Enterprise Intelligence Platform must integrate logs from numerous third-party vendors in order to perform various downstream tasks. However, vendor documentation is often unavailable at test time. It is either misplaced, mismatched, poorly formatted, or incomplete, which makes schema mapping challenging. We introduce a reinforcement learning agent that can self-improve without labeled examples or model weight updates. During inference, the agent: 1) Identifies ambiguous field-mapping attempts. 2) Generates targeted web-search queries to gather external evidence. 3) Applies a confidence-based reward to iteratively refine its mappings. To demonstrate this concept, we converted Microsoft Defender for Endpoint logs into a common schema. Our method increased mapping accuracy from 56.4%(LLM-only) to 72.73%(RAG) to 93.94% over 100 iterations using GPT-4o. At the same time, it reduced the number of low-confidence mappings requiring expert review by 85%. This new approach provides an evidence-driven, transparent method for solving future industry problems, paving the way for more robust, accountable, scalable, efficient, flexible, adaptable, and collaborative solutions.

To preserve data privacy, multi-party computation (MPC) enables executing Machine Learning (ML) algorithms on private data. However, MPC frameworks do not include optimized operations on sparse data. This absence makes them unsuitable for ML applications involving sparse data; e.g., recommender systems or genomics. Even in plaintext, such applications involve high-dimensional sparse data, that cannot be processed without sparsity-related optimizations due to prohibitively large memory requirements. Since matrix multiplication is a central building block of ML algorithms, our work proposes dedicated MPC algorithms to multiply secret-shared sparse matrices. Our sparse algorithms have several advantages over secure dense matrix multiplications (i.e., the classic multiplication). On the one hand, they avoid the memory issues caused by the "dense" data representation of dense multiplications. On the other hand, our algorithms can significantly reduce communication costs (up to $\times1000$) for realistic problem sizes. We validate our algorithms in two machine learning applications where dense matrix multiplications are impractical. Finally, we take inspiration from real-world sparse data properties to build 3 techniques minimizing the public knowledge necessary to secure sparse algorithms.

Machine Unlearning aims to remove specific data from trained models, addressing growing privacy and ethical concerns. We provide a theoretical analysis of a simple and widely used method - gradient ascent - used to reverse the influence of a specific data point without retraining from scratch. Leveraging the implicit bias of gradient descent towards solutions that satisfy the Karush-Kuhn-Tucker (KKT) conditions of a margin maximization problem, we quantify the quality of the unlearned model by evaluating how well it satisfies these conditions w.r.t. the retained data. To formalize this idea, we propose a new success criterion, termed \textbf{$(\epsilon, \delta, \tau)$-successful} unlearning, and show that, for both linear models and two-layer neural networks with high dimensional data, a properly scaled gradient-ascent step satisfies this criterion and yields a model that closely approximates the retrained solution on the retained data. We also show that gradient ascent performs successful unlearning while still preserving generalization in a synthetic Gaussian-mixture setting.

We experimentally demonstrate a new widespread read disturbance phenomenon, ColumnDisturb, in real commodity DRAM chips. By repeatedly opening or keeping a DRAM row (aggressor row) open, we show that it is possible to disturb DRAM cells through a DRAM column (i.e., bitline) and induce bitflips in DRAM cells sharing the same columns as the aggressor row (across multiple DRAM subarrays). With ColumnDisturb, the activation of a single row concurrently disturbs cells across as many as three subarrays (e.g., 3072 rows) as opposed to RowHammer/RowPress, which affect only a few neighboring rows of the aggressor row in a single subarray. We rigorously characterize ColumnDisturb and its characteristics under various operational conditions using 216 DDR4 and 4 HBM2 chips from three major manufacturers. Among our 27 key experimental observations, we highlight two major results and their implications. First, ColumnDisturb affects chips from all three major manufacturers and worsens as DRAM technology scales down to smaller node sizes (e.g., the minimum time to induce the first ColumnDisturb bitflip reduces by up to 5.06x). We observe that, in existing DRAM chips, ColumnDisturb induces bitflips within a standard DDR4 refresh window (e.g., in 63.6 ms) in multiple cells. We predict that, as DRAM technology node size reduces, ColumnDisturb would worsen in future DRAM chips, likely causing many more bitflips in the standard refresh window. Second, ColumnDisturb induces bitflips in many (up to 198x) more rows than retention failures. Therefore, ColumnDisturb has strong implications for retention-aware refresh mechanisms that leverage the heterogeneity in cell retention times: our detailed analyses show that ColumnDisturb greatly reduces the benefits of such mechanisms.