SECUTE 2026 – Preliminary Table of Contents

2nd International Workshop on Software Security Testing (SECUTE 2026)

Frontmatter

Papers

Using LLM Agents for Security Testing of Compiled Mobile Applications: A Case Study on a Real-World Compiled iOS App
Fabian Scherf and Florian Magin
(Fraunhofer SIT, Germany; ATHENE, Germany)
Security testing of compiled applications is crucial for organizations to assess the behavior and security of the software they deploy without relying on the vendor. In this paper, we present a case study that investigates the capability of autonomous Large Language Model (LLM) agents to detect security weaknesses and compliance risks in compiled code of a real-world iOS application. These application-layer vulnerabilities are often context-dependent: they arise from app logic patterns that violate security principles in ways that require a human-like understanding to recognize, making them resistant to detection via conventional static analysis methods.
We design and perform an evaluation on a real-world iOS app with several confirmed weaknesses that we discovered and reported to the vendor. We derive two task types: (1) directed tasks, where the system is given context on a security issue and must locate the relevant code, and (2) reporting tasks, where the model receives an open-ended task to analyze the app for security issues. We evaluate 20 model configurations across single-agent and multi-agent system architectures using open-weight LLMs.
Our evaluation shows that real-world, context‑dependent vulnerabilities and compliance risks, which are difficult to detect using conventional static analysis methods and usually rely on manual assessments, are reliably detectable using LLM agents operating on sufficiently reconstructed pseudo-code. Additionally, we observe that improved decompilation output is crucial to enable models to consistently solve the most complex tasks. We conclude that developers should use LLMs at the source-code level to counter adversaries applying the same techniques on decompiled code and research should advance decompilation support for the growing language diversity in the iOS ecosystem.
Article Search Artifacts Available Article: asews26secutemain-id1-p (type: Full Paper) doi:10.1145/3843781.3844621

Supplementary Material for the Paper: "Using LLM Agents for Security Testing of Compiled Mobile Applications" (doi:10.6084/m9.figshare.32955989.v2): The artifact contains supplementary material for the paper "Using LLM Agents for Security Testing of Compiled Mobile Applications", such as used LLM prompts, selected trajectories and additional evaluation data.

AutoDriver: Purely Source-Based Driver Generation for Standard Library Fuzzing Pipelines
Vincent Ahlrichs, Florian Kasten, Jonas Bogenberger, Dieter Schuster, and Julian Horsch
(Fraunhofer AISEC, Germany; Technical University of Munich, Germany)
Fuzzing libraries typically requires a fuzz driver that executes the library, as there is usually no distinct entry point. Writing these programs has been a complex manual task. In recent work, several approaches have been proposed to automatically generate fuzz drivers. However, these approaches either require extensive sample code using the library, rely on manually generated data, e.g., execution traces, or do not integrate into established fuzzing pipelines. Similarly, the quality of LLM-generated drivers may depend on the target’s representation in the model’s training data.
In this work, we present AutoDriver, an automated fuzz driver generator for C libraries. AutoDriver performs static analysis on the target’s source code to enrich the library headers with API semantics and uses these to automatically generate fuzz drivers that achieve broad and deep coverage during fuzzing with commonly deployed fuzzing engines, without requiring manual effort.
We evaluated AutoDriver on seven well-known libraries from diverse domains and benchmarked it against available manually written drivers. Against the best-performing manual drivers, the generated drivers improve coverage for four of seven libraries. Averaged over all libraries, they reach 11.8 and 23.2 percentage points higher line and function coverage, respectively. Our evaluation also confirms the ability of the generated drivers to reproduce real-world bugs, making AutoDriver a viable solution for library fuzzing.
Article Search Article: asews26secutemain-id3-p (type: Full Paper) doi:10.1145/3843781.3844623

Sentry: Towards a Cross-Language Dynamic Analysis Framework for Hybrid Android Applications
Jyoti Prakash, Mathias Ransgaard Yde, Miguel Enrique Campusano Araya, and Abhishek Tiwari
(University of Southern Denmark, Denmark)
Hybrid Android applications combine native components with web content rendered in a WebView. Security-relevant behaviour can therefore cross the Android–JavaScript boundary through external page loading, bridge method exposure, or browser-side operations. We present Sentry, a dynamic analysis framework that pairs na- tive instrumentation with a page-level JavaScript probe. Sentry records WebView-related events on both the native Android side and the JavaScript side, for a trace analyser to inspect. We evaluate Sentry on 18 benchmarks drawn from the IWandroid and Hybrid- Droid suites. The results show that Sentry distinguishes calls to the original native bridge from JavaScript-side bridge mutations and preserves the ordering of bridge exposure, native updates, and bridge invocations occur.
Article Search Article: asews26secutemain-id4-p (type: Short Paper (5 pages)) doi:10.1145/3843781.3844624

The Role of Prompt Patterns in LLM-Based Smart Contract Vulnerability Detection
Gerardo Iuliano and Dario Di Nucci
(University of Salerno, Italy)
Smart contracts are immutable programs deployed on blockchain platforms, and their vulnerabilities can cause severe financial losses. Therefore, several tools have been proposed to detect these vulnerabilities. Among them, recent research has increasingly leveraged Large Language Models (LLMs) relying on carefully crafted prompts. Meanwhile, prior work has identified several prompt patterns, such as persona roles, chain-of-thought reasoning, and multi-agent interactions. However, the sensitivity of LLM performance to prompt design raises questions about the reliability of existing evaluations.
This paper investigates how prompt patterns affect the effectiveness of LLMs in detecting smart contract vulnerabilities. Building on the zero-shot prompt used by Chen et al., which combines persona and chain-of-thought (CoT) patterns, we conducted an ablation study to measure the impact of removing specific patterns on detection accuracy. We further extended the original prompt with few-shot examples to evaluate potential performance gains over the baseline. Finally, we investigated the impact of prompt patterns on a reasoning model. The results show that LLMs without reasoning mechanisms obtain limited benefits from persona and CoT patterns, whereas few-shot prompting achieves the best precision-recall balance. In reasoning-enabled LLMs, simple task-definition prompts outperform persona and CoT patterns, whereas few-shot examples further improve precision and generalization.
Article Search Article: asews26secutemain-id7-p (type: Full Paper) doi:10.1145/3843781.3844627

Artifacts to Actors: Assessing the Maturity of Software Supply Chain Security Measurement Practices
Alexis Butler, Dan O'Keeffe, and Santanu Kumar Dash
(Royal Holloway University of London, UK; University of Surrey, UK)
Software supply chain attacks are an increasing threat. Industry reports from vendors, and open-source foundations play an important role in shaping how practitioners understand these risks and allocate their security testing budget. However, it remains unclear which aspects of supply chain security these reports measure and which they overlook. To address this gap, we conduct a systematic gray literature review of 37 reports published between 2021 and 2025, identified through Google searches in two countries and LLM-assisted discovery. Two reviewers independently screened the reports and extracted their metrics, classifying each by supply chain concept and measurement focus: Technical Artifact (TA), People & Practices (PP), or Practitioner Sentiment (PS). At the report level, 59.5% of reports contain a majority of TA metrics. Security vendors account for 46.9% of publishers and devote 66.2% of their metrics to TA, a pattern we describe as the “vendor lens effect.” We also find that several areas central to security testing practice, including security requirements specification and threat modeling, are absent or underrepresented. These findings indicate that industry measurement remains largely tool-driven, emphasizing artifact properties that can be readily automated. More holistic assessments will require practitioners and researchers to develop and validate complementary measures of people and practices.
Article Search Article: asews26secutemain-id8-p (type: Full Paper) doi:10.1145/3843781.3844628

proc time: 0.24