arXiv subjects

Yuekang Li

Publications and source records attributed to Yuekang Li.

At least 55 records · Page 3Linked to original sources

arXiv2025

Help or Hurdle? Rethinking Model Context Protocol-Augmented Large Language Models

The Model Context Protocol (MCP) enables large language models (LLMs) to access external resources on demand. While commonly assumed to enhance performance, how LLMs actually leverage this capability remains poorly understood. We introduce MCPGAUGE, the first comprehensive evaluation framework for probing LLM-MCP interactions along four key dimensions: proactivity (self-initiated tool use), compliance (adherence to tool-use instructions), effectiveness (task performance post-integration), and overhead (computational cost incurred). MCPGAUGE comprises a 160-prompt suite and 25 datasets spanning knowledge comprehension, general reasoning, and code generation. Our large-scale evaluation, spanning six commercial LLMs, 30 MCP tool suites, and both one- and two-turn interaction settings, comprises around 20,000 API calls and over USD 6,000 in computational cost. This comprehensive study reveals four key findings that challenge prevailing assumptions about the effectiveness of MCP integration. These insights highlight critical limitations in current AI-tool integration and position MCPGAUGE as a principled benchmark for advancing controllable, tool-augmented LLMs.

cs.AI↗

arXiv2025

"Pull or Not to Pull?'': Investigating Moral Biases in Leading Large Language Models Across Ethical Dilemmas

As large language models (LLMs) increasingly mediate ethically sensitive decisions, understanding their moral reasoning processes becomes imperative. This study presents a comprehensive empirical evaluation of 14 leading LLMs, both reasoning enabled and general purpose, across 27 diverse trolley problem scenarios, framed by ten moral philosophies, including utilitarianism, deontology, and altruism. Using a factorial prompting protocol, we elicited 3,780 binary decisions and natural language justifications, enabling analysis along axes of decisional assertiveness, explanation answer consistency, public moral alignment, and sensitivity to ethically irrelevant cues. Our findings reveal significant variability across ethical frames and model types: reasoning enhanced models demonstrate greater decisiveness and structured justifications, yet do not always align better with human consensus. Notably, "sweet zones" emerge in altruistic, fairness, and virtue ethics framings, where models achieve a balance of high intervention rates, low explanation conflict, and minimal divergence from aggregated human judgments. However, models diverge under frames emphasizing kinship, legality, or self interest, often producing ethically controversial outcomes. These patterns suggest that moral prompting is not only a behavioral modifier but also a diagnostic tool for uncovering latent alignment philosophies across providers. We advocate for moral reasoning to become a primary axis in LLM alignment, calling for standardized benchmarks that evaluate not just what LLMs decide, but how and why.

cs.CL↗

arXiv2025

CrossPL: Evaluating Large Language Models on Cross Programming Language Code Generation

As large language models (LLMs) become increasingly embedded in software engineering workflows, a critical capability remains underexplored: generating correct code that enables cross-programming-language (CPL) interoperability. This skill is essential for building complex systems that integrate components written in multiple languages via mechanisms like inter-process communication (IPC). To bridge this gap, we present CrossPL, the first benchmark designed to systematically evaluate LLMs' ability to generate CPL-interoperating code. CrossPL comprises 1,982 tasks centered around IPC, covering six widely-used programming languages and seven representative CPL techniques. We construct this benchmark by (i) analyzing 19,169 multi-language GitHub repositories using 156 hand-crafted finite state machines (FSMs), and (ii) developing an LLM-based pipeline that automatically extracts CPL code snippets, generates task instructions, and validates functional correctness. We evaluate 14 state-of-the-art general-purpose LLMs and 6 code-oriented LLMs released in the past three years on CrossPL via FSM-based validation. Results reveal that even the best-performing models struggle with CPL scenarios, underscoring the need for more targeted research in this space. Our benchmark and code are available at: https://anonymous.4open.science/r/crosspl-2814.

cs.SE↗

arXiv2025

It Only Gets Worse: Revisiting DL-Based Vulnerability Detectors from a Practical Perspective

With the growing threat of software vulnerabilities, deep learning (DL)-based detectors have gained popularity for vulnerability detection. However, doubts remain regarding their consistency within declared CWE ranges, real-world effectiveness, and applicability across scenarios. These issues may lead to unreliable detection, high false positives/negatives, and poor adaptability to emerging vulnerabilities. A comprehensive analysis is needed to uncover critical factors affecting detection and guide improvements in model design and deployment. In this paper, we present VulTegra, a novel evaluation framework that conducts a multidimensional comparison of scratch-trained and pre-trained-based DL models for vulnerability detection. VulTegra reveals that state-of-the-art (SOTA) detectors still suffer from low consistency, limited real-world capabilities, and scalability challenges. Contrary to common belief, pre-trained models are not consistently better than scratch-trained models but exhibit distinct strengths in specific contexts.Importantly, our study exposes the limitations of relying solely on CWE-based classification and identifies key factors that significantly affect model performance. Experimental results show that adjusting just one such factor consistently improves recall across all seven evaluated detectors, with six also achieving better F1 scores. Our findings provide deeper insights into model behavior and emphasize the need to consider both vulnerability types and inherent code features for effective detection.

cs.SE↗

arXiv2025

MMLU-ProX: A Multilingual Benchmark for Advanced Large Language Model Evaluation

Existing large language model (LLM) evaluation benchmarks primarily focus on English, while current multilingual tasks lack parallel questions that specifically assess cross-linguistic reasoning abilities. This dual limitation makes it challenging to comprehensively assess LLMs' performance in the multilingual setting. To fill this gap, we introduce MMLU-ProX, a comprehensive benchmark covering 29 languages, built on an English benchmark. Each language version consists of 11,829 identical questions, enabling direct cross-linguistic comparisons. Additionally, to meet efficient evaluation needs, we provide a lite version containing 658 questions per language. To ensure the high quality of MMLU-ProX, we employ a rigorous development process that involves multiple powerful LLMs for translation, followed by expert review to ensure accurate expression, consistent terminology, and cultural relevance. Building on this, we systematically evaluate 36 state-of-the-art LLMs, including reasoning-enhanced and multilingual-optimized LLMs. The results reveal significant disparities in the multilingual capabilities of LLMs: While they perform well in high-resource languages, their performance declines markedly in low-resource languages, with gaps of up to 24.3%. Through MMLU-ProX, we aim to advance the development of more inclusive AI systems and promote equitable access to technology across global contexts.

cs.CL↗

arXiv2025

Good News for Script Kiddies? Evaluating Large Language Models for Automated Exploit Generation

Large Language Models (LLMs) have demonstrated remarkable capabilities in code-related tasks, raising concerns about their potential for automated exploit generation (AEG). This paper presents the first systematic study on LLMs' effectiveness in AEG, evaluating both their cooperativeness and technical proficiency. To mitigate dataset bias, we introduce a benchmark with refactored versions of five software security labs. Additionally, we design an LLM-based attacker to systematically prompt LLMs for exploit generation. Our experiments reveal that GPT-4 and GPT-4o exhibit high cooperativeness, comparable to uncensored models, while Llama3 is the most resistant. However, no model successfully generates exploits for refactored labs, though GPT-4o's minimal errors highlight the potential for LLM-driven AEG advancements.

cs.CR↗

arXiv2025

A Rusty Link in the AI Supply Chain: Detecting Evil Configurations in Model Repositories

Recent advancements in large language models (LLMs) have spurred the development of diverse AI applications from code generation and video editing to text generation; however, AI supply chains such as Hugging Face, which host pretrained models and their associated configuration files contributed by the public, face significant security challenges; in particular, configuration files originally intended to set up models by specifying parameters and initial settings can be exploited to execute unauthorized code, yet research has largely overlooked their security compared to that of the models themselves; in this work, we present the first comprehensive study of malicious configurations on Hugging Face, identifying three attack scenarios (file, website, and repository operations) that expose inherent risks; to address these threats, we introduce CONFIGSCAN, an LLM-based tool that analyzes configuration files in the context of their associated runtime code and critical libraries, effectively detecting suspicious elements with low false positive rates and high accuracy; our extensive evaluation uncovers thousands of suspicious repositories and configuration files, underscoring the urgent need for enhanced security validation in AI model hosting platforms.

cs.CR↗

arXiv2025

ai.txt: A Domain-Specific Language for Guiding AI Interactions with the Internet

We introduce ai.txt, a novel domain-specific language (DSL) designed to explicitly regulate interactions between AI models, agents, and web content, addressing critical limitations of the widely adopted robots.txt standard. As AI increasingly engages with online materials for tasks such as training, summarization, and content modification, existing regulatory methods lack the necessary granularity and semantic expressiveness to ensure ethical and legal compliance. ai.txt extends traditional URL-based access controls by enabling precise element-level regulations and incorporating natural language instructions interpretable by AI systems. To facilitate practical deployment, we provide an integrated development environment with code autocompletion and automatic XML generation. Furthermore, we propose two compliance mechanisms: XML-based programmatic enforcement and natural language prompt integration, and demonstrate their effectiveness through preliminary experiments and case studies. Our approach aims to aid the governance of AI-Internet interactions, promoting responsible AI use in digital ecosystems.

cs.NI↗

arXiv2025

Demystifying RCE Vulnerabilities in LLM-Integrated Apps

LLMs show promise in transforming software development, with a growing interest in integrating them into more intelligent apps. Frameworks like LangChain aid LLM-integrated app development, offering code execution utility/APIs for custom actions. However, these capabilities theoretically introduce Remote Code Execution (RCE) vulnerabilities, enabling remote code execution through prompt injections. No prior research systematically investigates these frameworks' RCE vulnerabilities or their impact on applications and exploitation consequences. Therefore, there is a huge research gap in this field. In this study, we propose LLMSmith to detect, validate and exploit the RCE vulnerabilities in LLM-integrated frameworks and apps. To achieve this goal, we develop two novel techniques, including 1) a lightweight static analysis to examine LLM integration mechanisms, and construct call chains to identify RCE vulnerabilities in frameworks; 2) a systematical prompt-based exploitation method to verify and exploit the found vulnerabilities in LLM-integrated apps. This technique involves various strategies to control LLM outputs, trigger RCE vulnerabilities and launch subsequent attacks. Our research has uncovered a total of 20 vulnerabilities in 11 LLM-integrated frameworks, comprising 19 RCE vulnerabilities and 1 arbitrary file read/write vulnerability. Of these, 17 have been confirmed by the framework developers, with 11 vulnerabilities being assigned CVE IDs. For the 51 apps potentially affected by RCE, we successfully executed attacks on 17 apps, 16 of which are vulnerable to RCE and 1 to SQL injection. Furthermore, we conduct a comprehensive analysis of these vulnerabilities and construct practical attacks to demonstrate the hazards in reality. Last, we propose several mitigation measures for both framework and app developers to counteract such attacks.

cs.CR↗

arXiv2025

IllusionCAPTCHA: A CAPTCHA based on Visual Illusion

CAPTCHAs have long been essential tools for protecting applications from automated bots. Initially designed as simple questions to distinguish humans from bots, they have become increasingly complex to keep pace with the proliferation of CAPTCHA-cracking techniques employed by malicious actors. However, with the advent of advanced large language models (LLMs), the effectiveness of existing CAPTCHAs is now being undermined. To address this issue, we have conducted an empirical study to evaluate the performance of multimodal LLMs in solving CAPTCHAs and to assess how many attempts human users typically need to pass them. Our findings reveal that while LLMs can solve most CAPTCHAs, they struggle with those requiring complex reasoning type of CAPTCHA that also presents significant challenges for human users. Interestingly, our user study shows that the majority of human participants require a second attempt to pass these reasoning CAPTCHAs, a finding not reported in previous research. Based on empirical findings, we present IllusionCAPTCHA, a novel security mechanism employing the "Human-Easy but AI-Hard" paradigm. This new CAPTCHA employs visual illusions to create tasks that are intuitive for humans but highly confusing for AI models. Furthermore, we developed a structured, step-by-step method that generates misleading options, which particularly guide LLMs towards making incorrect choices and reduce their chances of successfully solving CAPTCHAs. Our evaluation shows that IllusionCAPTCHA can effectively deceive LLMs 100% of the time. Moreover, our structured design significantly increases the likelihood of AI errors when attempting to solve these challenges. Results from our user study indicate that 86.95% of participants successfully passed the CAPTCHA on their first attempt, outperforming other CAPTCHA systems.

cs.CR↗

arXiv2025

Detecting LLM Fact-conflicting Hallucinations Enhanced by Temporal-logic-based Reasoning

Large language models (LLMs) face the challenge of hallucinations -- outputs that seem coherent but are actually incorrect. A particularly damaging type is fact-conflicting hallucination (FCH), where generated content contradicts established facts. Addressing FCH presents three main challenges: 1) Automatically constructing and maintaining large-scale benchmark datasets is difficult and resource-intensive; 2) Generating complex and efficient test cases that the LLM has not been trained on -- especially those involving intricate temporal features -- is challenging, yet crucial for eliciting hallucinations; and 3) Validating the reasoning behind LLM outputs is inherently difficult, particularly with complex logical relationships, as it requires transparency in the model's decision-making process. This paper presents Drowzee, an innovative end-to-end metamorphic testing framework that utilizes temporal logic to identify fact-conflicting hallucinations (FCH) in large language models (LLMs). Drowzee builds a comprehensive factual knowledge base by crawling sources like Wikipedia and uses automated temporal-logic reasoning to convert this knowledge into a large, extensible set of test cases with ground truth answers. LLMs are tested using these cases through template-based prompts, which require them to generate both answers and reasoning steps. To validate the reasoning, we propose two semantic-aware oracles that compare the semantic structure of LLM outputs to the ground truths. Across nine LLMs in nine different knowledge domains, experimental results show that Drowzee effectively identifies rates of non-temporal-related hallucinations ranging from 24.7% to 59.8%, and rates of temporal-related hallucinations ranging from 16.7% to 39.2%.

cs.CL↗

arXiv2024

MiniScope: Automated UI Exploration and Privacy Inconsistency Detection of MiniApps via Two-phase Iterative Hybrid Analysis

The advent of MiniApps, operating within larger SuperApps, has revolutionized user experiences by offering a wide range of services without the need for individual app downloads. However, this convenience has raised significant privacy concerns, as these MiniApps often require access to sensitive data, potentially leading to privacy violations. Despite existing privacy regulations and platform guidelines, there is a lack of effective mechanisms to safeguard user privacy fully. To address this critical gap, we introduce MiniScope, a novel two-phase hybrid analysis approach, specifically designed for the MiniApp environment. This approach overcomes the limitations of existing static analysis techniques by incorporating UI transition states analysis, cross-package callback control flow resolution, and automated iterative UI exploration. This allows for a comprehensive understanding of MiniApps' privacy practices, addressing the unique challenges of sub-package loading and event-driven callbacks. Our empirical evaluation of over 120K MiniApps using MiniScope demonstrates its effectiveness in identifying privacy inconsistencies. The results reveal significant issues, with 5.7% of MiniApps over-collecting private data and 33.4% overclaiming data collection. We have responsibly disclosed our findings to 2,282 developers, receiving 44 acknowledgments. These findings emphasize the urgent need for more precise privacy monitoring systems and highlight the responsibility of SuperApp operators to enforce stricter privacy measures.

cs.CR↗

arXiv2024

TransferFuzz: Fuzzing with Historical Trace for Verifying Propagated Vulnerability Code

Code reuse in software development frequently facilitates the spread of vulnerabilities, making the scope of affected software in CVE reports imprecise. Traditional methods primarily focus on identifying reused vulnerability code within target software, yet they cannot verify if these vulnerabilities can be triggered in new software contexts. This limitation often results in false positives. In this paper, we introduce TransferFuzz, a novel vulnerability verification framework, to verify whether vulnerabilities propagated through code reuse can be triggered in new software. Innovatively, we collected runtime information during the execution or fuzzing of the basic binary (the vulnerable binary detailed in CVE reports). This process allowed us to extract historical traces, which proved instrumental in guiding the fuzzing process for the target binary (the new binary that reused the vulnerable function). TransferFuzz introduces a unique Key Bytes Guided Mutation strategy and a Nested Simulated Annealing algorithm, which transfers these historical traces to implement trace-guided fuzzing on the target binary, facilitating the accurate and efficient verification of the propagated vulnerability. Our evaluation, conducted on widely recognized datasets, shows that TransferFuzz can quickly validate vulnerabilities previously unverifiable with existing techniques. Its verification speed is 2.5 to 26.2 times faster than existing methods. Moreover, TransferFuzz has proven its effectiveness by expanding the impacted software scope for 15 vulnerabilities listed in CVE reports, increasing the number of affected binaries from 15 to 53. The datasets and source code used in this article are available at https://github.com/Siyuan-Li201/TransferFuzz.

cs.SE↗

arXiv2024

A Survey of Protocol Fuzzing

Communication protocols form the bedrock of our interconnected world, yet vulnerabilities within their implementations pose significant security threats. Recent developments have seen a surge in fuzzing-based research dedicated to uncovering these vulnerabilities within protocol implementations. However, there still lacks a systematic overview of protocol fuzzing for answering the essential questions such as what the unique challenges are, how existing works solve them, etc. To bridge this gap, we conducted a comprehensive investigation of related works from both academia and industry. Our study includes a detailed summary of the specific challenges in protocol fuzzing, and provides a systematic categorization and overview of existing research efforts. Furthermore, we explore and discuss potential future research directions in protocol fuzzing. This survey serves as a foundational guideline for researchers and practitioners in the field.

cs.CR↗

arXiv2024

Drowzee: Metamorphic Testing for Fact-Conflicting Hallucination Detection in Large Language Models

Large language models (LLMs) have transformed the landscape of language processing, yet struggle with significant challenges in terms of security, privacy, and the generation of seemingly coherent but factually inaccurate outputs, commonly referred to as hallucinations. Among these challenges, one particularly pressing issue is Fact-Conflicting Hallucination (FCH), where LLMs generate content that directly contradicts established facts. Tackling FCH poses a formidable task due to two primary obstacles: Firstly, automating the construction and updating of benchmark datasets is challenging, as current methods rely on static benchmarks that don't cover the diverse range of FCH scenarios. Secondly, validating LLM outputs' reasoning process is inherently complex, especially with intricate logical relations involved. In addressing these obstacles, we propose an innovative approach leveraging logic programming to enhance metamorphic testing for detecting Fact-Conflicting Hallucinations (FCH). Our method gathers data from sources like Wikipedia, expands it with logical reasoning to create diverse test cases, assesses LLMs through structured prompts, and validates their coherence using semantic-aware assessment mechanisms. Our method generates test cases and detects hallucinations across six different LLMs spanning nine domains, revealing hallucination rates ranging from 24.7% to 59.8%. Key observations indicate that LLMs encounter challenges, particularly with temporal concepts, handling out-of-distribution knowledge, and exhibiting deficiencies in logical reasoning capabilities. The outcomes underscore the efficacy of logic-based test cases generated by our tool in both triggering and identifying hallucinations. These findings underscore the imperative for ongoing collaborative endeavors within the community to detect and address LLM hallucinations.

cs.SE↗

arXiv2024

Image-Based Geolocation Using Large Vision-Language Models

Geolocation is now a vital aspect of modern life, offering numerous benefits but also presenting serious privacy concerns. The advent of large vision-language models (LVLMs) with advanced image-processing capabilities introduces new risks, as these models can inadvertently reveal sensitive geolocation information. This paper presents the first in-depth study analyzing the challenges posed by traditional deep learning and LVLM-based geolocation methods. Our findings reveal that LVLMs can accurately determine geolocations from images, even without explicit geographic training. To address these challenges, we introduce \tool{}, an innovative framework that significantly enhances image-based geolocation accuracy. \tool{} employs a systematic chain-of-thought (CoT) approach, mimicking human geoguessing strategies by carefully analyzing visual and contextual cues such as vehicle types, architectural styles, natural landscapes, and cultural elements. Extensive testing on a dataset of 50,000 ground-truth data points shows that \tool{} outperforms both traditional models and human benchmarks in accuracy. It achieves an impressive average score of 4550.5 in the GeoGuessr game, with an 85.37\% win rate, and delivers highly precise geolocation predictions, with the closest distances as accurate as 0.3 km. Furthermore, our study highlights issues related to dataset integrity, leading to the creation of a more robust dataset and a refined framework that leverages LVLMs' cognitive capabilities to improve geolocation precision. These findings underscore \tool{}'s superior ability to interpret complex visual data, the urgent need to address emerging security vulnerabilities posed by LVLMs, and the importance of responsible AI development to ensure user privacy protection.

cs.CR↗

arXiv2024

How Effective Are They? Exploring Large Language Model Based Fuzz Driver Generation

LLM-based (Large Language Model) fuzz driver generation is a promising research area. Unlike traditional program analysis-based method, this text-based approach is more general and capable of harnessing a variety of API usage information, resulting in code that is friendly for human readers. However, there is still a lack of understanding regarding the fundamental issues on this direction, such as its effectiveness and potential challenges. To bridge this gap, we conducted the first in-depth study targeting the important issues of using LLMs to generate effective fuzz drivers. Our study features a curated dataset with 86 fuzz driver generation questions from 30 widely-used C projects. Six prompting strategies are designed and tested across five state-of-the-art LLMs with five different temperature settings. In total, our study evaluated 736,430 generated fuzz drivers, with 0.85 billion token costs ($8,000+ charged tokens). Additionally, we compared the LLM-generated drivers against those utilized in industry, conducting extensive fuzzing experiments (3.75 CPU-year). Our study uncovered that: - While LLM-based fuzz driver generation is a promising direction, it still encounters several obstacles towards practical applications; - LLMs face difficulties in generating effective fuzz drivers for APIs with intricate specifics. Three featured design choices of prompt strategies can be beneficial: issuing repeat queries, querying with examples, and employing an iterative querying process; - While LLM-generated drivers can yield fuzzing outcomes that are on par with those used in the industry, there are substantial opportunities for enhancement, such as extending contained API usage, or integrating semantic oracles to facilitate logical bug detection. Our insights have been implemented to improve the OSS-Fuzz-Gen project, facilitating practical fuzz driver generation in industry.

cs.CR↗

arXiv2024

Continuous Embedding Attacks via Clipped Inputs in Jailbreaking Large Language Models

Security concerns for large language models (LLMs) have recently escalated, focusing on thwarting jailbreaking attempts in discrete prompts. However, the exploration of jailbreak vulnerabilities arising from continuous embeddings has been limited, as prior approaches primarily involved appending discrete or continuous suffixes to inputs. Our study presents a novel channel for conducting direct attacks on LLM inputs, eliminating the need for suffix addition or specific questions provided that the desired output is predefined. We additionally observe that extensive iterations often lead to overfitting, characterized by repetition in the output. To counteract this, we propose a simple yet effective strategy named CLIP. Our experiments show that for an input length of 40 at iteration 1000, applying CLIP improves the ASR from 62% to 83%

cs.CR↗