内容库 / 论文

arXiv:2608.03700 · 评测基准

When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills

当 Agent 学会成为你：Persona Skill 中隐私泄漏、冒充风险与防御的基准评测

When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills

类型：arxiv
标识：2608.03700
链接：https://arxiv.org/abs/2608.03700
主分类：evaluation
形态：benchmark
被引：0
被引来源：Semantic Scholar
S2被引：0
影响力被引：0
TLDR：AntiSkillBench is introduced, an end-to-end benchmark for evaluating risks and defenses across the persona-skill pipeline, and experiments show that persona-skill risks persist across agent backbones and distillation protocols, extending from explicit attributes to communication styles and personality traits.
副分类：agent
待LLM分类：否
标题中文：当 Agent 学会成为你：Persona Skill 中隐私泄漏、冒充风险与防御的基准评测
TLDR中文：提出 AntiSkillBench，一个端到端的基准，用于评估 persona-skill 流水线中的风险与防御；实验表明 persona-skill 风险在不同的 agent backbone 和蒸馏协议下持续存在，从显式属性扩展到沟通风格与个性特征。
来源文件：
/inbox/tom/_candidates/2026-08-05-agent-rag-longcontext-candidates.json
[S2 enrich]

相关 · 按语义匹配

笔记深度精读 · SoK: Agentic Retrieval-Augmented Generation（arXiv:2603.07379）

笔记 SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents — 短精读

笔记 Tom 文献雷达 · Agent / RAG / 长上下文 · 2026-08-22

仓库 JeffffffFu/Awesome-Differential-Privacy-and-Meachine-Learning

仓库 sisinflab/adversarial-recommender-systems-survey

笔记 Tom 文献雷达 · Agent RAG LongContext · 2026-07-07 14:30

arXiv

2608.03700

TLDR

提出 AntiSkillBench，一个端到端的基准，用于评估 persona-skill 流水线中的风险与防御；实验表明 persona-skill 风险在不同的 agent backbone 和蒸馏协议下持续存在，从显式属性扩展到沟通风格与个性特征。AntiSkillBench is introduced, an end-to-end benchmark for evaluating risks and defenses across the persona-skill pipeline, and experiments show that persona-skill risks persist across agent backbones and distillation protocols, extending from explicit attributes to communication styles and personality traits.

被引 · S2

0

S2 0

影响力 0

相关解读

arXiv ↗ 中文全文