Agent Safety / More Autonomous Agents

• SlideDeck: W10.2-team4-agent-safety
• Version: current
• Lead team: team-4

Safety Agent

In this session, our readings cover:

Required Readings:

AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents

• Maksym Andriushchenko, Alexandra Souly, Mateusz Dziemian, Derek Duenas, Maxwell Lin, Justin Wang, Dan Hendrycks, Andy Zou, Zico Kolter, Matt Fredrikson, Eric Winsor, Jerome Wynne, Yarin Gal, Xander Davies
• [Submitted on 11 Oct 2024 (v1), last revised 14 Oct 2024 (this version, v2)]
• The robustness of LLMs to jailbreak attacks, where users design prompts to circumvent safety measures and misuse model capabilities, has been studied primarily for LLMs acting as simple chatbots. Meanwhile, LLM agents – which use external tools and can execute multi-stage tasks – may pose a greater risk if misused, but their robustness remains underexplored. To facilitate research on LLM agent misuse, we propose a new benchmark called AgentHarm. The benchmark includes a diverse set of 110 explicitly malicious agent tasks (440 with augmentations), covering 11 harm categories including fraud, cybercrime, and harassment. In addition to measuring whether models refuse harmful agentic requests, scoring well on AgentHarm requires jailbroken agents to maintain their capabilities following an attack to complete a multi-step task. We evaluate a range of leading LLMs, and find (1) leading LLMs are surprisingly compliant with malicious agent requests without jailbreaking, (2) simple universal jailbreak templates can be adapted to effectively jailbreak agents, and (3) these jailbreaks enable coherent and malicious multi-step agent behavior and retain model capabilities. To enable simple and reliable evaluation of attacks and defenses for LLM-based agents, we publicly release AgentHarm at this https URL.

UDora: A Unified Red Teaming Framework against LLM Agents by Dynamically Hijacking Their Own Reasoning

[Submitted on 28 Feb 2025] Jiawei Zhang, Shuang Yang, Bo Li Large Language Model (LLM) agents equipped with external tools have become increasingly powerful for handling complex tasks such as web shopping, automated email replies, and financial trading. However, these advancements also amplify the risks of adversarial attacks, particularly when LLM agents can access sensitive external functionalities. Moreover, because LLM agents engage in extensive reasoning or planning before executing final actions, manipulating them into performing targeted malicious actions or invoking specific tools remains a significant challenge. Consequently, directly embedding adversarial strings in malicious instructions or injecting malicious prompts into tool interactions has become less effective against modern LLM agents. In this work, we present UDora, a unified red teaming framework designed for LLM Agents that dynamically leverages the agent’s own reasoning processes to compel it toward malicious behavior. Specifically, UDora first samples the model’s reasoning for the given task, then automatically identifies multiple optimal positions within these reasoning traces to insert targeted perturbations. Subsequently, it uses the modified reasoning as the objective to optimize the adversarial strings. By iteratively applying this process, the LLM agent will then be induced to undertake designated malicious actions or to invoke specific malicious tools. Our approach demonstrates superior effectiveness compared to existing methods across three LLM agent datasets.

More Readings:

The Emerged Security and Privacy of LLM Agent: A Survey with Case Studies

• [Submitted on 28 Jul 2024]
• Feng He, Tianqing Zhu, Dayong Ye, Bo Liu, Wanlei Zhou, Philip S. Yu Inspired by the rapid development of Large Language Models (LLMs), LLM agents have evolved to perform complex tasks. LLM agents are now extensively applied across various domains, handling vast amounts of data to interact with humans and execute tasks. The widespread applications of LLM agents demonstrate their significant commercial value; however, they also expose security and privacy vulnerabilities. At the current stage, comprehensive research on the security and privacy of LLM agents is highly needed. This survey aims to provide a comprehensive overview of the newly emerged privacy and security issues faced by LLM agents. We begin by introducing the fundamental knowledge of LLM agents, followed by a categorization and analysis of the threats. We then discuss the impacts of these threats on humans, environment, and other agents. Subsequently, we review existing defensive strategies, and finally explore future trends. Additionally, the survey incorporates diverse case studies to facilitate a more accessible understanding. By highlighting these critical security and privacy issues, the survey seeks to stimulate future research towards enhancing the security and privacy of LLM agents, thereby increasing their reliability and trustworthiness in future applications.

Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models

• Yongshuo Zong, Ondrej Bohdal, Tingyang Yu, Yongxin Yang, Timothy Hospedales
• [Submitted on 3 Feb 2024 (v1), last revised 17 Jun 2024 (this version, v2)]
• Current vision large language models (VLLMs) exhibit remarkable capabilities yet are prone to generate harmful content and are vulnerable to even the simplest jailbreaking attacks. Our initial analysis finds that this is due to the presence of harmful data during vision-language instruction fine-tuning, and that VLLM fine-tuning can cause forgetting of safety alignment previously learned by the underpinning LLM. To address this issue, we first curate a vision-language safe instruction-following dataset VLGuard covering various harmful categories. Our experiments demonstrate that integrating this dataset into standard vision-language fine-tuning or utilizing it for post-hoc fine-tuning effectively safety aligns VLLMs. This alignment is achieved with minimal impact on, or even enhancement of, the models’ helpfulness. The versatility of our safety fine-tuning dataset makes it a valuable resource for safety-testing existing VLLMs, training new models or safeguarding pre-trained VLLMs. Empirical results demonstrate that fine-tuned VLLMs effectively reject unsafe instructions and substantially reduce the success rates of several black-box adversarial attacks, which approach zero in many cases. The code and dataset are available at this https URL.

Unique Security and Privacy Threats of Large Language Model: A Comprehensive Survey

• [Submitted on 12 Jun 2024 (v1), last revised 18 Jun 2024 (this version, v2)]
• Shang Wang, Tianqing Zhu, Bo Liu, Ming Ding, Xu Guo, Dayong Ye, Wanlei Zhou, Philip S. Yu
• With the rapid development of artificial intelligence, large language models (LLMs) have made remarkable advancements in natural language processing. These models are trained on vast datasets to exhibit powerful language understanding and generation capabilities across various applications, including machine translation, chatbots, and agents. However, LLMs have revealed a variety of privacy and security issues throughout their life cycle, drawing significant academic and industrial attention. Moreover, the risks faced by LLMs differ significantly from those encountered by traditional language models. Given that current surveys lack a clear taxonomy of unique threat models across diverse scenarios, we emphasize the unique privacy and security threats associated with five specific scenarios: pre-training, fine-tuning, retrieval-augmented generation systems, deployment, and LLM-based agents. Addressing the characteristics of each risk, this survey outlines potential threats and countermeasures. Research on attack and defense situations can offer feasible research directions, enabling more areas to benefit from LLMs.

Large Language Model Safety: A Holistic Survey

• Dan Shi, Tianhao Shen, Yufei Huang, Zhigen Li, Yongqi Leng, Renren Jin, Chuang Liu, Xinwei Wu, Zishan Guo, Linhao Yu, Ling Shi, Bojian Jiang, Deyi Xiong
• [Submitted on 23 Dec 2024]
• The rapid development and deployment of large language models (LLMs) have introduced a new frontier in artificial intelligence, marked by unprecedented capabilities in natural language understanding and generation. However, the increasing integration of these models into critical applications raises substantial safety concerns, necessitating a thorough examination of their potential risks and associated mitigation strategies. This survey provides a comprehensive overview of the current landscape of LLM safety, covering four major categories: value misalignment, robustness to adversarial attacks, misuse, and autonomous AI risks. In addition to the comprehensive review of the mitigation methodologies and evaluation resources on these four aspects, we further explore four topics related to LLM safety: the safety implications of LLM agents, the role of interpretability in enhancing LLM safety, the technology roadmaps proposed and abided by a list of AI companies and institutes for LLM safety, and AI governance aimed at LLM safety with discussions on international cooperation, policy proposals, and prospective regulatory directions. Our findings underscore the necessity for a proactive, multifaceted approach to LLM safety, emphasizing the integration of technical solutions, ethical considerations, and robust governance frameworks. This survey is intended to serve as a foundational resource for academy researchers, industry practitioners, and policymakers, offering insights into the challenges and opportunities associated with the safe integration of LLMs into society. Ultimately, it seeks to contribute to the safe and beneficial development of LLMs, aligning with the overarching goal of harnessing AI for societal advancement and well-being. A curated list of related papers has been publicly available at this https URL.

MobileSafetyBench: Evaluating Safety of Autonomous Agents in Mobile Device Control

• https://arxiv.org/pdf/2410.17520
• [Submitted on 23 Oct 2024 (v1), last revised 10 Dec 2024 (this version, v2)]
• Juyong Lee, Dongyoon Hahm, June Suk Choi, W. Bradley Knox, Kimin Lee
• Autonomous agents powered by large language models (LLMs) show promising potential in assistive tasks across various domains, including mobile device control. As these agents interact directly with personal information and device settings, ensuring their safe and reliable behavior is crucial to prevent undesirable outcomes. However, no benchmark exists for standardized evaluation of the safety of mobile device-control agents. In this work, we introduce MobileSafetyBench, a benchmark designed to evaluate the safety of device-control agents within a realistic mobile environment based on Android emulators. We develop a diverse set of tasks involving interactions with various mobile applications, including messaging and banking applications, challenging agents with managing risks encompassing misuse and negative side effects. These tasks include tests to evaluate the safety of agents in daily scenarios as well as their robustness against indirect prompt injection attacks. Our experiments demonstrate that baseline agents, based on state-of-the-art LLMs, often fail to effectively prevent harm while performing the tasks. To mitigate these safety concerns, we propose a prompting method that encourages agents to prioritize safety considerations. While this method shows promise in promoting safer behaviors, there is still considerable room for improvement to fully earn user trust. This highlights the urgent need for continued research to develop more robust safety mechanisms in mobile environments. We open-source our benchmark at: this https URL.

Privacy-Preserving Large Language Models: Mechanisms, Applications, and Future Directions

• Guoshenghui Zhao, Eric Song
• [Submitted on 9 Dec 2024]
• The rapid advancement of large language models (LLMs) has revolutionized natural language processing, enabling applications in diverse domains such as healthcare, finance and education. However, the growing reliance on extensive data for training and inference has raised significant privacy concerns, ranging from data leakage to adversarial attacks. This survey comprehensively explores the landscape of privacy-preserving mechanisms tailored for LLMs, including differential privacy, federated learning, cryptographic protocols, and trusted execution environments. We examine their efficacy in addressing key privacy challenges, such as membership inference and model inversion attacks, while balancing trade-offs between privacy and model utility. Furthermore, we analyze privacy-preserving applications of LLMs in privacy-sensitive domains, highlighting successful implementations and inherent limitations. Finally, this survey identifies emerging research directions, emphasizing the need for novel frameworks that integrate privacy by design into the lifecycle of LLMs. By synthesizing state-of-the-art approaches and future trends, this paper provides a foundation for developing robust, privacy-preserving large language models that safeguard sensitive information without compromising performance.