Cryptography & Security

Mateusz Dziemian, Maxwell Lin, Xiaohan Fu et al. · eth-zurich

This addresses a critical security threat for users of AI agents in high-stakes settings, revealing fundamental weaknesses in current models.

Xinhao Deng, Yixiang Zhang, Jiaqing Wu et al.

This addresses security risks for users and developers of autonomous LLM agents, but it is incremental as it builds on existing threat analysis frameworks.

Chuan Guo, Juan Felipe Ceron Uribe, Sicheng Zhu et al.

This addresses security vulnerabilities in frontier LLMs for developers and users, though it is incremental as it builds on existing reinforcement learning and dataset methods.

Yu He, Haozhe Zhu, Yiming Li et al.

This addresses a critical security vulnerability in LLM agents for users deploying them in untrusted environments, offering a novel defense paradigm that is resilient to adaptive attacks.

Yuanhe Zhang, Xinyue Wang, Zhican Chen et al.

This is an incremental survey that addresses resource efficiency issues for LLM providers and users.

Davis Brown, Mahdi Sabbaghi, Luze Sun et al.

For AI safety researchers, this work highlights a practical attack vector and provides benchmarks to evaluate defenses against covert misuse.

Chenlong Yin, Runpeng Geng, Yanting Wang et al.

This addresses security risks for real-world LLM applications, particularly autonomous agents, by providing a systematic evaluation method, though it is incremental as it builds on existing RL and red-teaming approaches.

Yihao Zhang, Zeming Wei, Xiaokun Luan et al.

This addresses critical security risks for users of interconnected multi-agent systems, exposing vulnerabilities that could lead to autonomous attacks without attacker intervention.

Pengfei He, Yue Xing, Juanhui Li et al.

For researchers and practitioners building LLM-MAS, this work provides foundational groundwork for security analysis, but it is primarily a position paper without empirical results.

Juhee Kim, Xiaoyuan Liu, Zhun Wang et al.

It addresses security problems for developers and researchers in AI agent systems, but as a survey, it is incremental in synthesizing existing knowledge rather than proposing new methods.

Quanchen Zou, Moyang Chen, Zonghao Ying et al.

This work addresses a systemic flaw in LVLM safety for users relying on secure AI systems, representing a novel attack paradigm rather than an incremental improvement.

Meenatchi Sundaram Muthu Selva Annamalai, Emiliano De Cristofaro, Peter Kairouz

This work exposes critical privacy risks in widely used AI platforms, highlighting insufficient protections for user data, and is incremental as it tests existing claims rather than proposing new defenses.

Vincent Siu, Jingxuan He, Kyle Montgomery et al.

This work provides a foundational framework for improving security in LLM agents, though it is incremental as it systematizes existing concepts rather than introducing new methods.

Zijian Ling, Pingyi Hu, Xiuyong Gao et al.

This addresses a critical security problem for users of speech-driven LLMs by demonstrating practical, black-box attacks that are perceptually undetectable, though it is incremental in applying known acoustic techniques to a new domain.

Kai Wang, Biaojie Zeng, Zeming Wei et al.

This addresses safety risks for developers and users of multi-agent systems, though it appears incremental as it builds on existing standards like OWASP.

James Flemings, Ren Yi, Octavian Suciu et al.

This addresses privacy concerns for users of personal LLM agents by improving decision accuracy, though it is incremental as it builds on existing logic and LLM integration approaches.

Mihai Christodorescu, Earlence Fernandes, Ashish Hooda et al.

For AI safety researchers and developers, it reframes agent security from a model-centric to a systems-centric approach, highlighting the insufficiency of model robustness alone.

Ada Chen, Yongjiang Wu, Junyuan Zhang et al. · pku, tencent-ai

For researchers and practitioners developing or deploying LLM-based autonomous agents, this survey systematizes emerging safety and security risks, offering a comprehensive reference.

Tom Sander, Hongyan Chang, Tomáš Souček et al.

For LLM developers and deployers, TextSeal provides a practical, distortion-free watermarking method that is robust to dilution and supports serving optimizations, addressing the need for provenance and distillation protection.

Yuan Xin, Yixuan Weng, Minjun Zhu et al.

For academic peer review systems using LLMs, this work provides a dynamic defense against adversarial manipulation, though it is an incremental step in adversarial robustness.