Shuai Wang

This paper introduces a novel framework, the Reasoning Safety Monitor, to detect and prevent logical inconsistencies and adversarial manipulations within the internal reasoning steps of large language models, establishing reasoning safety as a critical security dimension.

The paper introduces ReproMIA, a novel and efficient framework that uses model reprogramming to proactively amplify and detect latent privacy leakage for Membership Inference Attacks (MIAs), significantly outperforming state-of-the-art methods, especially in low False Positive Rate regimes.

The paper independently stress-tests Claude Code's auto mode permission system using a deliberately ambiguous benchmark, finding that its true false negative rate is significantly higher than reported, particularly due to unmonitored file edits.

ClawLess introduces a formally verified security framework that enforces fine-grained policies on autonomous AI agents, mitigating risks associated with their ability to run code and retrieve information.

ZK-Value introduces a practical, scalable zero-knowledge system for calculating data valuations (Shapley values) in data marketplaces, significantly reducing proving time while maintaining high accuracy.

SkillScope introduces a graph-based framework to enforce fine-grained least-privilege in LLM Agent Skills, significantly reducing over-privileged actions while maintaining task functionality.

This paper re-evaluates prompt-injection attacks in realistic RAG settings, finding that most prior attack methods fail to reach the generator, and that current attacks are easily detectable.

The paper introduces SURE, a unified framework designed to standardize and improve the comparability and reproducibility of evaluations for advanced speech understanding models.