Rui Liu

The paper proposes FRA-Attack, a frequency-domain regularization method, to significantly improve the transferability of adversarial attacks against closed-source Multimodal Large Language Models (MLLMs).

The paper introduces a multilingual benchmark (MentalMap) to test if LLMs build internal spatial world models from text, finding a universal 'L3 reasoning cliff' suggesting that text-only working memory is the primary bottleneck.

This paper introduces the concept of 'Sleeper Attack,' demonstrating that adversarial content can persist across multiple interactions with an LLM agent, posing a more subtle and difficult-to-detect safety threat than single-interaction attacks.

The paper introduces AgentDoG 1.5, a lightweight and scalable alignment framework that significantly improves AI agent safety and security for complex, open-world agentic scenarios.

UI-KOBE is a framework that enhances lightweight mobile GUI agents by integrating reusable, app-specific knowledge graphs, allowing them to perform complex tasks efficiently on-device without relying on large vision-language models.

The paper introduces AgentDoG 1.5, a lightweight and scalable alignment framework that significantly improves AI agent safety and security for complex open-world agent deployments.

COLLEAGUE.SKILL introduces an automated system that distills heterogeneous traces of human expertise and role-specific knowledge into portable, inspectable, and usable AI skill packages.

The paper introduces the Data-centric Reasoning Compiler (DCRC), a novel data-driven framework that enhances financial QA systems by compiling user queries and retrieved documents into verifiable, executable programs to prevent numerical hallucinations.

The paper introduces PrivacyPeek, a new benchmark that audits the acquisition stage of LLM-based agents to show that unnecessary and sensitive data acquisition is a widespread and critical privacy vulnerability.

The paper introduces PrivacyPeek, a new benchmark that audits the acquisition stage of LLM-based agents to demonstrate that unnecessary acquisition of sensitive data is a widespread and critical privacy vulnerability.

The paper introduces TaskWeave, a hierarchical agentic framework that successfully simulates long-horizon organizational dynamics by treating coordination as a memory-centered problem, demonstrating that structured memory is key to reliable LLM-based simulations.

The paper introduces HLL, a benchmark that tests if multimodal agents can successfully substitute for human verification (like CAPTCHA) in complex, real-world workflows, finding that current agents are still brittle and fail under realistic conditions.

The paper introduces X-Stream, a new benchmark for multi-stream video understanding, and finds that current state-of-the-art MLLMs perform poorly when required to process multiple concurrent video streams.

This survey provides a systematic framework and taxonomy for evidence tracing and execution provenance in LLM agents, addressing the difficulty of verifying and auditing complex agent behaviors.