Michael Backes

Michael Backes

Auto-generated profile, based on publicly available information

Michael Backes

CISPA Helmholtz Center for Information Security, Saarbrücken, Germany

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 December 2024
Verified
Conference Paper
Open access
XSXinyue Shen
Xinyue Shen
ZCZeyuan Chen
Zeyuan Chen
MBMichael Backes
Michael Backes
YSYun Shen
Yun Shen
YZYang Zhang
Yang Zhang
The misuse of large language models (LLMs) has drawn significant attention from the general public and LLM vendors. One particular type of adversarial prompt, known as jailbreak prompt, has emerged as the main attack vector to bypass the safeguards and elicit harmful content from LLMs. In this paper, employing our new framework JailbreakHub, we conduct a comprehensive analysis of 1,405 jailbreak prompts spanning from December 2022 to December 2023. We identify 131 jailbreak communities and discover unique characteristics of jailbreak prompts and their major attack strategies, such as prompt injection and privilege escalation. We also observe that jailbreak prompts increasingly shift from online Web communities to prompt-aggregation websites and 28 user accounts have consistently optimized jailbreak prompts over 100 days. To assess the potential harm caused by jailbreak prompts, we create a question set comprising 107,250 samples across 13 forbidden scenarios. Leveraging this dataset, our experiments on six popular LLMs show that their safeguards cannot adequately defend jailbreak prompts in all scenarios. Particularly, we identify five highly effective jailbreak prompts that achieve 0.95 attack success rates on ChatGPT (GPT-3.5) and GPT-4, and the earliest one has persisted online for over 240 days. We hope that our study can facilitate the research community and LLM vendors in promoting safer and regulated LLMs.

2 December 2024
Verified
Conference Paper
Open access
XHXinlei He
Xinlei He
XSXinyue Shen
Xinyue Shen
ZCZeyuan Chen
Zeyuan Chen
MBMichael Backes
Michael Backes
YZYang Zhang
Yang Zhang
Nowadays, powerful large language models (LLMs) such as ChatGPT have demonstrated revolutionary power in a variety of natural language processing (NLP) tasks such as text classification, sentiment analysis, language translation, and question-answering. Consequently, the detection of machine-generated texts (MGTs) is becoming increasingly crucial as LLMs become more advanced and prevalent. These models have the ability to generate human-like language, making it challenging to discern whether a text is authored by a human or a machine. This raises concerns regarding authenticity, accountability, and potential bias. However, existing methods for detecting MGTs are evaluated using different model architectures, datasets, and experimental settings, resulting in a lack of a comprehensive evaluation framework that encompasses various methodologies. Furthermore, it remains unclear how existing detection methods would perform against powerful LLMs. In this paper, we fill this gap by proposing the first benchmark framework for MGT detection against powerful LLMs, named MGTBench. Extensive evaluations on public datasets with curated texts generated by various powerful LLMs such as ChatGPT-turbo and Claude demonstrate the effectiveness of different detection methods. Our ablation study shows that a larger number of words in general leads to better performance and most detection methods can achieve similar performance with much fewer training samples. Additionally, our findings reveal that metric-based/model-based detection methods exhibit better transferability across different LLMs/datasets. Furthermore, we delve into a more challenging task: text attribution, where the goal is to identify the originating model of a given text, i.e., whether it is a specific LLM or authored by a human. Our findings indicate that the model-based detection methods still perform well in the text attribution task. To investigate the robustness of different detection methods, we consider three adversarial attacks, namely paraphrasing, random spacing, and adversarial perturbations. We discover that these attacks can significantly diminish detection effectiveness, underscoring the critical need for the development of more robust detection methods. We envision that MGTBench will serve as a benchmark tool to accelerate future investigations involving the evaluation of powerful MGT detection methods on their respective datasets and the development of more advanced MGT detection methods.

15 November 2023
Verified
Conference Paper
Open access
YQYiting Qu
Yiting Qu
XSXinyue Shen
Xinyue Shen
XHXinlei He
Xinlei He
MBMichael Backes
Michael Backes
SZSavvas Zannettou
Savvas Zannettou
YZYang Zhang
Yang Zhang
State-of-the-art Text-to-Image models like Stable Diffusion and DALLE\cdot2 are revolutionizing how people generate visual content. At the same time, society has serious concerns about how adversaries can exploit such models to generate problematic or unsafe images. In this work, we focus on demystifying the generation of unsafe images and hateful memes from Text-to-Image models. We first construct a typology of unsafe images consisting of five categories (sexually explicit, violent, disturbing, hateful, and political). Then, we assess the proportion of unsafe images generated by four advanced Text-to-Image models using four prompt datasets. We find that Text-to-Image models can generate a substantial percentage of unsafe images; across four models and four prompt datasets, 14.56% of all generated images are unsafe. When comparing the four Text-to-Image models, we find different risk levels, with Stable Diffusion being the most prone to generating unsafe content (18.92% of all generated images are unsafe). Given Stable Diffusion's tendency to generate more unsafe content, we evaluate its potential to generate hateful meme variants if exploited by an adversary to attack a specific individual or community. We employ three image editing methods, DreamBooth, Textual Inversion, and SDEdit, which are supported by Stable Diffusion to generate variants. Our evaluation result shows that 24% of the generated images using DreamBooth are hateful meme variants that present the features of the original hateful meme and the target individual/community; these generated images are comparable to hateful meme variants collected from the real world. Overall, our results demonstrate that the danger of large-scale generation of unsafe images is imminent. We discuss several mitigating measures, such as curating training data, regulating prompts, and implementing safety filters, and encourage better safeguard tools to be developed to prevent unsafe generation.1 Our code is available at https://github.com/YitingQu/unsafe-diffusion.

19 November 2025
Verified
Conference Paper
Open access
YQYiting Qu
Yiting Qu
XSXinyue Shen
Xinyue Shen
YWYixin Wu
Yixin Wu
MBMichael Backes
Michael Backes
SZSavvas Zannettou
Savvas Zannettou
YZYang Zhang
Yang Zhang

2 December 2024
Verified
Conference Paper
Open access
RWRui Wen
Rui Wen
ZLZheng Li
Zheng Li
MBMichael Backes
Michael Backes
YZYang Zhang
Yang Zhang
Adapting Large Language Models (LLMs) to specific tasks introduces concerns about computational efficiency, prompting an exploration of efficient methods such as In-Context Learning (ICL). However, the vulnerability of ICL to privacy attacks under realistic assumptions remains largely unexplored. In this work, we present the first membership inference attack tailored for ICL, relying solely on generated texts without their associated probabilities. We propose four attack strategies tailored to various constrained scenarios and conduct extensive experiments on four popular large language models. Empirical results show that our attacks can accurately determine membership status in most cases, e.g., 95% accuracy advantage against LLaMA, indicating that the associated risks are much higher than those shown by existing probability-based attacks. Additionally, we propose a hybrid attack that synthesizes the strengths of the aforementioned strategies, achieving an accuracy advantage of over 95% in most cases. Furthermore, we investigate three potential defenses targeting data, instruction, and output. Results demonstrate combining defenses from orthogonal dimensions significantly reduces privacy leakage and offers enhanced privacy assurances.

7 November 2022
Verified
Conference Paper
Open access
TNTrung Tin Nguyen
Trung Tin Nguyen
MBMichael Backes
Michael Backes
BSBen Stock
Ben Stock
Adopted in May 2018, the European Union's General Data Protection Regulation (GDPR) requires the consent for processing users' personal data to be freely given, specific, informed, and unambiguous. While prior work has shown that this often is not given through automated network traffic analysis, no research has systematically studied how consent notices are currently implemented and whether they conform to GDPR in mobile apps. To close this research gap, we perform the first large-scale study into consent notices for third-party tracking in Android apps to understand the current practices and the current state of GDPR's consent violations. Specifically, we propose a mostly automated and scalable approach to identify the currently implemented consent notices and apply it to a set of 239,381 Android apps. As a result, we recognize four widely implemented mechanisms to interact with the consent user interfaces from 13,082 apps. We then develop a tool that automatically detects users' personal data sent out to the Internet with different consent conditions based on the identified mechanisms. Doing so, we find 30,160 apps do not even attempt to implement consent notices for sharing users' personal data with third-party data controllers, which mandate explicit consent under GDPR. In contrast, out of 13,082 apps implemented consent notices, we identify 2,688 (20.54%) apps violate at least one of the GDPR consent requirements, such as trying to deceive users into accepting all data sharing or even continuously transmitting data when users have explicitly opted out. To allow developers to address the problems, we send emails to notify affected developers and gather insights from their responses. Our study shows the urgent need for more transparent processing of personal data and supporting developers in this endeavor to comply with legislation, ensuring users can make free and informed choices regarding their data.