用因果推断框架审计合成数据中的幻影与隐私泄露风险，理论严谨且实用。
arXiv:2606.16952v1 Announce Type: cross Abstract: The rapid adoption of generative AI and Large Language Models (LLMs) has spurred interest in synthet…
提出LoRA-MINT方法，专为审计领域适应型LLM的训练数据，有效追踪数据来源与隐私风险。
arXiv:2606.06946v1 Announce Type: cross Abstract: We present LoRA-MINT, a new methodology for Membership Inference Test (MINT) applied to recent Large…
用项目反应理论深度剖析LLM基准测试，揭示评估偏差与模型能力真实度量。
arXiv:2605.30504v1 Announce Type: new Abstract: LLM benchmark labels are frozen at release and silently propagated into downstream benchmarks, errors …
基于信息操纵理论审计大模型欺骗行为，新方法直击AI安全盲区。
arXiv:2605.19270v1 Announce Type: new Abstract: Large language models can deceive by subtly manipulating truthful information -- omitting key facts, s…