News
Next Seminar on 20.11.2024
Written on 13.11.2024 21:08 by Xinyi Xu
Dear All,
The next seminar(s) will take place on 2024-11-20 at 14:30 (Session A) and 14:00 (Session B).
Session A: (14:30 - 15:00, 15:00 - 15:30)
Chun Ngai Li, Abdullah Alfurjani
https://cispa-de.zoom.us/j/96786205841?pwd=M3FOQ3dSczRabDNLb3F1czVXVUpvdz09
Meeting-ID: 967 8620 5841
Password: BT!u5=
Session B: (14:00 - 14:30, 14:30 - 15:00, 15:00 - 15:30)
Julian Rederlechner, Yannick Schording, Robin Wiesen
https://cispa-de.zoom-x.de/j/66136901453?pwd=YVBSZU9peUpvUlk4bWp3MDR4cGlUUT09
Meeting-ID: 661 3690 1453
Password: sxHhzA004}
Session A
14:30 - 15:00
Speaker: Chun Ngai Li
Type of Talk: Master Intro
Advisor: Thorsten Holz, Bhupendra Acharya
Title: Exploring the Cybersecurity Threats in LLM-Powered Apps: Malicious Code Generation and Regulatory Challenges
Research Area: RA5: Empirical and Behavioural Security
Abstract: This thesis explores the cybersecurity threats by absuing the Large Language Model (LLM)-powered apps, focusing on their malicious code generation capabilities, the challenges and concerns in regulating these misuses. With the expansion of LLM apps across sectors, their potential misuse for generating harmful outputs, such as phishing emails and websites, has become a critical concern. This study investigates the vulnerabilities in current LLM-powered apps, particularly those lacking sufficient safeguards, which can be exploited for malicious code generation. By analyzing various of LLM apps using jailbreaking techniques, and their output quality, this research aims to assess the resilience of these applications against misuse. This study also examines the regulatory and developer policies needed to mitigate these threats and highlights the societal implications if these issues are not addressed. The results of the study will provide insights for strengthening cybersecurity defenses for LLM-driven technologies and advocate for stronger oversight mechanisms.
15:00 - 15:30
Speaker: Abdullah Alfurjani
Type of Talk: Master Intro
Advisor: Thorsten Holz
Title: Fingerprinting Attributes Independence Measurement
Research Area: RA5: Empirical and Behavioural Security
Abstract: The objective of this research is to systematically evaluate the independence and uniqueness of various web fingerprinting attributes by conducting a comprehensive measurement study. This study will focus on fingerprinting APIs available on the web, including those related to Audio, Canvas, and JavaScript Floating Point operations, to assess their ability to uniquely identify users across diverse environments. We aim to explore how these attributes perform under different combinations of browsers, devices, and operating systems and to determine which attributes are most effective in generating distinct, reliable user fingerprints. To achieve this, we will conduct an extensive data collection and measurement campaign, gathering real-world fingerprints from users operating in various contexts. This includes different browser versions, operating systems, and hardware setups, allowing us to study how fingerprinting attributes perform across a wide range of scenarios. We will measure the relative uniqueness of each attribute and quantify the extent to which they contribute to creating distinct user profiles. This study addresses a critical gap in existing research by shifting the focus from isolated assessments of fingerprinting attributes, as seen in prior studies [6], [7], [11], [18], to a comprehensive analysis of how these attributes interact with one another. While previous research has primarily evaluated individual fingerprinting techniques, our approach considers the complex interplay between multiple attributes across various contexts. This allows for a more nuanced understanding of which combinations yield the most robust user identification. Furthermore, this research introduces a novel framework that ranks fingerprinting attributes not only by their individual effectiveness but also by their independence and ability to uniquely identify users when combined with other attributes. This dual focus on interaction and ranking sets our framework apart from earlier studies, providing a more holistic tool for understanding and improving fingerprinting techniques.
Session B
14:00 - 14:30
Speaker: Julian Rederlechner
Type of Talk: Bachelor Intro
Advisor: Ali Abbasi
Title: Spot the Diff-erence: Investigation of bsdiff
Research Area: RA3: Threat Detection and Defenses
Abstract: In an age where efficient software updates are crucial, especially for IoT devices, smartphones with limited connectivity and even vehicles, small and reliable over-the-air (OTA) updates have become an important topic. In this talk, we will focus on the aspect of "minimizing data transmission". We will present bsdiff, an efficient binary diffing algorithm originally developed to create compact software patches. Its early version, bsdiff4, set a standard for generating minimal patches that optimize update distribution. Its successor, bsdiff6, promises smaller patch sizes, but is still largely unexplored and unpublished. Our research aims to explore the structure and benefits of bsdiff6, and ultimately provide a modern Rust implementation. This project will not only shed light on the capabilities of bsdiff6, but also provide a baseline implementation and comprehensive documentation that will contribute to OTA solutions for networked devices in various industries.
14:30 - 15:00
Speaker: Yannick Schording
Type of Talk: Master Final
Advisor: Dominic Steinhoefel
Title: Specification-Based Testing with JSON Schemas
Research Area: RA3: Threat Detection and Defenses
Abstract: In recent years, the JSON data format has become one of the most popular formats for data interchange via the internet, especially for communication between API endpoints. Since the services that provide these endpoints often handle sensitive data, it is crucial that they work as intended and do not contain any bugs that could be abused for malicious purposes. Fuzzing is one of the techniques that can be used to make sure that this is the case. By generating numerous diverse inputs and feeding them to these systems, it is possible to discover the inputs that trigger such bugs. While generating completely random inputs might already discover some bugs, most of them will not adhere to the syntax or semantics expected by the tested software. They are rejected early by the software and thus cannot reach deeper parts of its code. A popular approach to prevent this is fuzzing based on the specification of the input language expected by the software. The most prominent specification format for JSON is JSON schema. Schemas are written as JSON objects themselves and define the structure other JSON objects should follow. In this thesis, we develop a fuzzing tool that automatically generates JSON data which adheres to the syntactic and semantic rules defined by the schema. It first translates the schema to a grammar and a set of ISLa constraints, and then uses the ISLa solver to produce valid inputs for it. To prove the capabilities of our tool, we compare the quality of its inputs to ones produced by the popular JSON Schema Faker library and test some popular software applications with it.
15:00 - 15:30
Speaker: Robin Wiesen
Type of Talk: Bachelor Final
Advisor: Sven Bugiel
Title: Selective Permissions for Android's SDK Runtime
Research Area: RA4: Secure Mobile and Autonomous Systems
Abstract: A frequently criticized aspect of Android’s security concept is that third-party libraries are executed within the host app’s sandbox and thus inherit all of its privileges. This gives them access to substantially more sensitive resources than necessary, which jeopardizes the security and privacy of users. In response, Android 13 introduced the SDK Runtime as an option to confine untrusted code in its own sandbox with separate privileges. However, the current design is primarily geared towards advertising SDKs and imposes rigid restrictions on the isolated libraries, such as a fixed set of permissions. As this significantly limits the applicability of the SDK Runtime, it is questionable how much the offered potential is actually used. The aim of this bachelor thesis is to develop a solution that enables code in the SDK Runtime to selectively request additional permissions. To this end, we build an application-layer extension where the host app takes on the role of a proxy and requests permissions and data on behalf of the SDK. The library enforces a security policy for requests from the SDK Runtime, effectively enabling flexible permissions without OS modifications. In our prototype, we demonstrate the functionality of this approach exemplarily for location and contact data. Although subsequent performance measurements reveal relatively high overhead in some cases, it demonstrates the feasibility of using the SDK Runtime as the basis for flexible privilege separation. Extending its scope beyond advertising libraries facilitates the implementation of a modular, least-privilege app architecture. At the same time, supporting a more fine-grained access control policy without impairing user experience raises usability challenges that represent an interesting area for future research.