arXiv subjects

Joanna C. S. Santos

Publications and source records attributed to Joanna C. S. Santos.

At least 19 recordsLinked to original sources

arXiv2026

Security in the Age of AI Teammates: An Empirical Study of Agentic Pull Requests on GitHub

Autonomous coding agents are increasingly deployed as AI teammates in modern software engineering, independently authoring pull requests (PRs) that modify production code at scale. This study aims to systematically characterize how autonomous coding agents contribute to software security in practice, how these security-related contributions are reviewed and accepted, and which observable signals are associated with PR rejection. We conduct a large-scale empirical analysis of agent-authored PRs using the AIDev dataset, comprising of over 33,000 curated PRs from popular GitHub repositories. Security-relevant PRs are identified using a keyword filtering strategy, followed by manual validation, resulting in 1,293 confirmed security-related agentic-PRs. We then analyze prevalence, acceptance outcomes, and review latency across autonomous agents, programming ecosystems, and types of code changes. Moreover, we apply qualitative open coding to identify recurring security-related actions and underlying intents, and examine review metadata to identify early signals associated with PR rejection. Security-related Agentic-PRs constitute a meaningful share of agent activity (approximately 4\%). Rather than focusing solely on narrow vulnerability fixes, agents most frequently perform supportive security hardening activities, including testing, documentation, configuration, and improved error handling. Compared to non-security PRs, security-related Agentic-PRs exhibit lower merge rates and longer review latency, reflecting heightened human scrutiny, with variation across agents and programming ecosystems. PR rejection is more strongly associated with PR complexity and verbosity than with explicit security topics.

cs.CR↗

arXiv2026

An Empirical Study of Security Calibration in Large Language Models for Code

Large Language Models (LLMs) are rapidly transforming software development, yet their use in security-critical contexts raises a key question: do models know when their generated code is insecure? This property, known as calibration, measures whether a model's confidence aligns with the true correctness of its outputs. We present the first large-scale empirical study of security calibration in LLM-generated code. We evaluate GPT-4o-mini, Gemini-2.0-Flash, and Qwen3-Coder-Next across multiple temperature settings on two complementary benchmarks: self-contained security tasks and multi-language repository-level contexts. Our results suggest that overconfidence is prevalent across the evaluated LLMs. Functional calibration is consistently worse than security calibration, suggesting that models estimate security outcomes more reliably than functional correctness, potentially because functional correctness depends on complex execution behavior. We also examine whether calibration-guided automated repair can help remediate vulnerabilities in LLM-generated code, finding only limited improvements while frequently introducing functional regressions. Moreover, we study different mitigation strategies for reducing False Trust, where models assign high confidence to vulnerable code. The results show that although architectural gating improves calibration on controlled benchmarks, calibration deteriorates in realistic repository-level settings, increasing the risk of high-confidence vulnerable outputs.

cs.SE↗

arXiv2026

Automated Detection of Configuration-Specific Security Vulnerabilities via Patch Analysis

We study how security patches in highly configurable C/C++ systems map onto the space of compile-time variants. We formalize the Vulnerability Impact Condition (VIC) - a Boolean predicate over configuration options that denotes all variants that contained the original flaw - and introduce PatchLens, a purely static technique that recovers VICs by aligning AST-level patch hunks with source-level presence conditions and resolving file inclusion via lightweight build system analysis. Evaluating PatchLens on 1,192 Linux kernel, 289 FFmpeg, and 100 PHP patches, we compute precise, human-readable VICs without the need to compile any system variant. The resulting predicates are compact (avg. 1.84 variables for Linux, 3.23 for FFmpeg, 1.04 for PHP) and show that only a small fraction of vulnerabilities are system-wide, which carry higher CVSS scores; meanwhile, CVE texts almost never encode the required options ($\approx$ 1% average recall), motivating automated enrichment of CVE descriptions with VICs. PatchLens and the accompanying dataset enable immediate applications in CI (variant-aware triage and test selection), targeted sampling and fuzzing, and feature risk scoring, offering a scalable, explainable path to vulnerability assessment in highly configurable software.

cs.SE↗

arXiv2026

Characterizing and Modeling the GitHub Security Advisories Review Pipeline

GitHub Security Advisories (GHSA) have become a central component of open-source vulnerability disclosure and are widely used by developers and security tools. A distinctive feature of GHSA is that only a fraction of advisories are reviewed by GitHub, while the mechanisms associated with this review process remain poorly understood. In this paper, we conduct a large-scale empirical study of the GHSA review processes, analyzing over 288,000 advisories spanning 2019-2025. We characterize which advisories are more likely to be reviewed, quantify review delays, and identify two distinct review-latency regimes: a fast path dominated by GitHub Repository Advisories (GRAs) and a slow path dominated by NVD-first advisories. We further develop a queueing model that accounts for this dichotomy based on the structure of the advisory processing pipeline.

cs.CR↗

arXiv2026

CodeGuard: Improving LLM Guardrails in CS Education

Large language models (LLMs) are increasingly embedded in Computer Science (CS) classrooms to automate code generation, feedback, and assessment. However, their susceptibility to adversarial or ill-intentioned prompts threatens student learning and academic integrity. To cope with this important issue, we evaluate existing off-the-shelf LLMs in handling unsafe and irrelevant prompts within the domain of CS education. We identify important shortcomings in existing LLM guardrails which motivates us to propose CodeGuard, a comprehensive guardrail framework for educational AI systems. CodeGuard includes (i) a first-of-its-kind taxonomy for classifying prompts; (ii) the CodeGuard dataset, a collection of 8,000 prompts spanning the taxonomy; and (iii) PromptShield, a lightweight sentence-encoder model fine-tuned to detect unsafe prompts in real time. Experiments show that PromptShield achieves 0.93 F1 score, surpassing existing guardrail methods. Additionally, further experimentation reveals that CodeGuard reduces potentially harmful or policy-violating code completions by 30-65% without degrading performance on legitimate educational tasks. The code, datasets, and evaluation scripts are made freely available to the community.

cs.CY↗

arXiv2026

An Empirical Study on Remote Code Execution in Machine Learning Model Hosting Ecosystems

Model-sharing platforms, such as Hugging Face, ModelScope, and OpenCSG, have become central to modern machine learning development, enabling developers to share, load, and fine-tune pre-trained models with minimal effort. However, the flexibility of these ecosystems introduces a critical security concern: the execution of untrusted code during model loading (i.e., via trust_remote_code or trust_repo). In this work, we conduct the first large-scale empirical study of custom model loading practices across five major model-sharing platforms to assess their prevalence, associated risks, and developer perceptions. We first quantify the frequency with which models require custom code to function and identify those that execute arbitrary Python files during loading. We then apply three complementary static analysis tools: Bandit, CodeQL, and Semgrep, to detect security smells and potential vulnerabilities, categorizing our findings by CWE identifiers to provide a standardized risk taxonomy. We also use YARA to identify malicious patterns and payload signatures. In parallel, we systematically analyze the documentation, API design, and safety mechanisms of each platform to understand their mitigation strategies and enforcement levels. Finally, we conduct a qualitative analysis of over 600 developer discussions from GitHub, Hugging Face, and PyTorch Hub forums, as well as Stack Overflow, to capture community concerns and misconceptions regarding security and usability. Our findings reveal widespread reliance on unsafe defaults, uneven security enforcement across platforms, and persistent confusion among developers about the implications of executing remote code. We conclude with actionable recommendations for designing safer model-sharing infrastructures and striking a balance between usability and security in future AI ecosystems.

cs.SE↗

arXiv2025

Assessing the Software Security Comprehension of Large Language Models

Large language models (LLMs) are increasingly used in software development, but their level of software security expertise remains unclear. This work systematically evaluates the security comprehension of five leading LLMs: GPT-4o-Mini, GPT-5-Mini, Gemini-2.5-Flash, Llama-3.1, and Qwen-2.5, using Blooms Taxonomy as a framework. We assess six cognitive dimensions: remembering, understanding, applying, analyzing, evaluating, and creating. Our methodology integrates diverse datasets, including curated multiple-choice questions, vulnerable code snippets (SALLM), course assessments from an Introduction to Software Security course, real-world case studies (XBOW), and project-based creation tasks from a Secure Software Engineering course. Results show that while LLMs perform well on lower-level cognitive tasks such as recalling facts and identifying known vulnerabilities, their performance degrades significantly on higher-order tasks that require reasoning, architectural evaluation, and secure system creation. Beyond reporting aggregate accuracy, we introduce a software security knowledge boundary that identifies the highest cognitive level at which a model consistently maintains reliable performance. In addition, we identify 51 recurring misconception patterns exhibited by LLMs across Blooms levels.

cs.SE↗

arXiv2025

Large Language Models for Software Engineering: A Reproducibility Crisis

Reproducibility is a cornerstone of scientific progress, yet its state in large language model (LLM)-based software engineering (SE) research remains poorly understood. This paper presents the first large-scale, empirical study of reproducibility practices in LLM-for-SE research. We systematically mined and analyzed 640 papers published between 2017 and 2025 across premier software engineering, machine learning, and natural language processing venues, extracting structured metadata from publications, repositories, and documentation. Guided by four research questions, we examine (i) the prevalence of reproducibility smells, (ii) how reproducibility has evolved over time, (iii) whether artifact evaluation badges reliably reflect reproducibility quality, and (iv) how publication venues influence transparency practices. Using a taxonomy of seven smell categories: Code and Execution, Data, Documentation, Environment and Tooling, Versioning, Model, and Access and Legal, we manually annotated all papers and associated artifacts. Our analysis reveals persistent gaps in artifact availability, environment specification, versioning rigor, and documentation clarity, despite modest improvements in recent years and increased adoption of artifact evaluation processes at top SE venues. Notably, we find that badges often signal artifact presence but do not consistently guarantee execution fidelity or long-term reproducibility. Motivated by these findings, we provide actionable recommendations to mitigate reproducibility smells and introduce a Reproducibility Maturity Model (RMM) to move beyond binary artifact certification toward multi-dimensional, progressive evaluation of reproducibility rigor.

cs.SE↗

arXiv2025

Attacking interpretable NLP systems

Studies have shown that machine learning systems are vulnerable to adversarial examples in theory and practice. Where previous attacks have focused mainly on visual models that exploit the difference between human and machine perception, text-based models have also fallen victim to these attacks. However, these attacks often fail to maintain the semantic meaning of the text and similarity. This paper introduces AdvChar, a black-box attack on Interpretable Natural Language Processing Systems, designed to mislead the classifier while keeping the interpretation similar to benign inputs, thus exploiting trust in system transparency. AdvChar achieves this by making less noticeable modifications to text input, forcing the deep learning classifier to make incorrect predictions and preserve the original interpretation. We use an interpretation-focused scoring approach to determine the most critical tokens that, when changed, can cause the classifier to misclassify the input. We apply simple character-level modifications to measure the importance of tokens, minimizing the difference between the original and new text while generating adversarial interpretations similar to benign ones. We thoroughly evaluated AdvChar by testing it against seven NLP models and three interpretation models using benchmark datasets for the classification task. Our experiments show that AdvChar can significantly reduce the prediction accuracy of current deep learning models by altering just two characters on average in input samples.

cs.CR↗

arXiv2025

A Survey of Source Code Representations for Machine Learning-Based Cybersecurity Tasks

Machine learning techniques for cybersecurity-related software engineering tasks are becoming increasingly popular. The representation of source code is a key portion of the technique that can impact the way the model is able to learn the features of the source code. With an increasing number of these techniques being developed, it is valuable to see the current state of the field to better understand what exists and what is not there yet. This article presents a study of these existing machine learning based approaches and demonstrates what type of representations were used for different cybersecurity tasks and programming languages. Additionally, we study what types of models are used with different representations. We have found that graph-based representations are the most popular category of representation, and tokenizers and Abstract Syntax Trees (ASTs) are the two most popular representations overall (e.g., AST and tokenizers are the representations with the highest count of papers, whereas graph-based representations is the category with the highest count of papers). We also found that the most popular cybersecurity task is vulnerability detection, and the language that is covered by the most techniques is C. Finally, we found that sequence-based models are the most popular category of models, and Support Vector Machines are the most popular model overall.

cs.LG↗

arXiv2025

An Empirical Study of Safetensors' Usage Trends and Developers' Perceptions

Developers are sharing pre-trained Machine Learning (ML) models through a variety of model sharing platforms, such as Hugging Face, in an effort to make ML development more collaborative. To share the models, they must first be serialized. While there are many methods of serialization in Python, most of them are unsafe. To tame this insecurity, Hugging Face released safetensors as a way to mitigate the threats posed by unsafe serialization formats. In this context, this paper investigates developer's shifts towards using safetensors on Hugging Face in an effort to understand security practices in the ML development community, as well as how developers react to new methods of serialization. Our results find that more developers are adopting safetensors, and many safetensor adoptions were made by automated conversions of existing models by Hugging Face's conversion tool. We also found, however, that a majority of developers ignore the conversion tool's pull requests, and that while many developers are facing issues with using safetensors, they are eager to learn about and adapt the format.

cs.SE↗

arXiv2024

MojoBench: Language Modeling and Benchmarks for Mojo

The recently introduced Mojo programming language (PL) by Modular, has received significant attention in the scientific community due to its claimed significant speed boost over Python. Despite advancements in code Large Language Models (LLMs) across various PLs, Mojo remains unexplored in this context. To address this gap, we introduce MojoBench, the first framework for Mojo code generation. MojoBench includes HumanEval-Mojo, a benchmark dataset designed for evaluating code LLMs on Mojo, and Mojo-Coder, the first LLM pretrained and finetuned for Mojo code generation, which supports instructions in 5 natural languages (NLs). Our results show that Mojo-Coder achieves a 30-35% performance improvement over leading models like GPT-4o and Claude-3.5-Sonnet. Furthermore, we provide insights into LLM behavior with underrepresented and unseen PLs, offering potential strategies for enhancing model adaptability. MojoBench contributes to our understanding of LLM capabilities and limitations in emerging programming paradigms fostering more robust code generation systems.

cs.CL↗

arXiv2024

Large Language Models in Computer Science Education: A Systematic Literature Review

Large language models (LLMs) are becoming increasingly better at a wide range of Natural Language Processing tasks (NLP), such as text generation and understanding. Recently, these models have extended their capabilities to coding tasks, bridging the gap between natural languages (NL) and programming languages (PL). Foundational models such as the Generative Pre-trained Transformer (GPT) and LLaMA series have set strong baseline performances in various NL and PL tasks. Additionally, several models have been fine-tuned specifically for code generation, showing significant improvements in code-related applications. Both foundational and fine-tuned models are increasingly used in education, helping students write, debug, and understand code. We present a comprehensive systematic literature review to examine the impact of LLMs in computer science and computer engineering education. We analyze their effectiveness in enhancing the learning experience, supporting personalized education, and aiding educators in curriculum development. We address five research questions to uncover insights into how LLMs contribute to educational outcomes, identify challenges, and suggest directions for future research.

cs.LG↗

arXiv2024

A Large-Scale Exploit Instrumentation Study of AI/ML Supply Chain Attacks in Hugging Face Models

The development of machine learning (ML) techniques has led to ample opportunities for developers to develop and deploy their own models. Hugging Face serves as an open source platform where developers can share and download other models in an effort to make ML development more collaborative. In order for models to be shared, they first need to be serialized. Certain Python serialization methods are considered unsafe, as they are vulnerable to object injection. This paper investigates the pervasiveness of these unsafe serialization methods across Hugging Face, and demonstrates through an exploitation approach, that models using unsafe serialization methods can be exploited and shared, creating an unsafe environment for ML developers. We investigate to what extent Hugging Face is able to flag repositories and files using unsafe serialization methods, and develop a technique to detect malicious models. Our results show that Hugging Face is home to a wide range of potentially vulnerable models.

cs.CR↗

arXiv2024

The Fault in our Stars: Quality Assessment of Code Generation Benchmarks

Large Language Models (LLMs) are gaining popularity among software engineers. A crucial aspect of developing effective code generation LLMs is to evaluate these models using a robust benchmark. Evaluation benchmarks with quality issues can provide a false sense of performance. In this work, we conduct the first-of-its-kind study of the quality of prompts within benchmarks used to compare the performance of different code generation models. To conduct this study, we analyzed 3,566 prompts from 9 code generation benchmarks to identify quality issues in them. We also investigated whether fixing the identified quality issues in the benchmarks' prompts affects a model's performance. We also studied memorization issues of the evaluation dataset, which can put into question a benchmark's trustworthiness. We found that code generation evaluation benchmarks mainly focused on Python and coding exercises and had very limited contextual dependencies to challenge the model. These datasets and the developers' prompts suffer from quality issues like spelling and grammatical errors, unclear sentences to express developers' intent, and not using proper documentation style. Fixing all these issues in the benchmarks can lead to a better performance for Python code generation, but not a significant improvement was observed for Java code generation. We also found evidence that GPT-3.5-Turbo and CodeGen-2.5 models may have data contamination issues.

cs.SE↗

arXiv2024

SALLM: Security Assessment of Generated Code

With the growing popularity of Large Language Models (LLMs) in software engineers' daily practices, it is important to ensure that the code generated by these tools is not only functionally correct but also free of vulnerabilities. Although LLMs can help developers to be more productive, prior empirical studies have shown that LLMs can generate insecure code. There are two contributing factors to the insecure code generation. First, existing datasets used to evaluate LLMs do not adequately represent genuine software engineering tasks sensitive to security. Instead, they are often based on competitive programming challenges or classroom-type coding tasks. In real-world applications, the code produced is integrated into larger codebases, introducing potential security risks. Second, existing evaluation metrics primarily focus on the functional correctness of the generated code while ignoring security considerations. Therefore, in this paper, we described SALLM, a framework to benchmark LLMs' abilities to generate secure code systematically. This framework has three major components: a novel dataset of security-centric Python prompts, configurable assessment techniques to evaluate the generated code, and novel metrics to evaluate the models' performance from the perspective of secure code generation.

cs.SE↗

arXiv2024

Seneca: Taint-Based Call Graph Construction for Java Object Deserialization

Object serialization and deserialization are widely used for storing and preserving objects in files, memory, or database as well as for transporting them across machines, enabling remote interaction among processes and many more. This mechanism relies on reflection, a dynamic language that introduces serious challenges for static analyses. Current state-of-the-art call graph construction algorithms do not fully support object serialization/deserialization, i.e., they are unable to uncover the callback methods that are invoked when objects are serialized and deserialized. Since call graphs are a core data structure for multiple types of analysis (e.g., vulnerability detection), an appropriate analysis cannot be performed since the call graph does not capture hidden (vulnerable) paths that occur via callback methods. In this paper, we present Seneca, an approach for handling serialization with improved soundness in the context of call graph construction. Our approach relies on taint analysis and API modeling to construct sound call graphs. We evaluated our approach with respect to soundness, precision, performance, and usefulness in detecting untrusted object deserialization vulnerabilities. Our results show that Seneca can create sound call graphs with respect to serialization features. The resulting call graphs do not incur significant runtime overhead and were shown to be useful for performing identification of vulnerable paths caused by untrusted object deserialization.

cs.SE↗

arXiv2024

FRANC: A Lightweight Framework for High-Quality Code Generation

In recent years, the use of automated source code generation utilizing transformer-based generative models has expanded, and these models can generate functional code according to the requirements of the developers. However, recent research revealed that these automatically generated source codes can contain vulnerabilities and other quality issues. Despite researchers' and practitioners' attempts to enhance code generation models, retraining and fine-tuning large language models is time-consuming and resource-intensive. Thus, we describe FRANC, a lightweight framework for recommending more secure and high-quality source code derived from transformer-based code generation models. FRANC includes a static filter to make the generated code compilable with heuristics and a quality-aware ranker to sort the code snippets based on a quality score. Moreover, the framework uses prompt engineering to fix persistent quality issues. We evaluated the framework with five Python and Java code generation models and six prompt datasets, including a newly created one in this work (SOEval). The static filter improves 9% to 46% Java suggestions and 10% to 43% Python suggestions regarding compilability. The average improvement over the NDCG@10 score for the ranking system is 0.0763, and the repairing techniques repair the highest 80% of prompts. FRANC takes, on average, 1.98 seconds for Java; for Python, it takes 0.08 seconds.

cs.SE↗