#privacy leaks

10 stories · 0 in 7 days · 2 in 30 days

AI safety 1 CIPL 1 DeepSeek-R1-32B 1 FedAvg 1 FLTrust 1 GPT-5 1 LLM 1 agents 1

Stories, products and papers tagged "privacy leaks", newest first.

Oct 21

Industry Policy02:21

Chinese contentMeta 暂停 Muse 聊天分享功能，研究人员发现 6800 多条对话链接可公开访问

Meta 的 Muse 聊天分享链接被研究人员用公共 DNS 扒出 6800 多条，都挂在公网上，Meta 只能先停掉这个功能。

#Meta #Muse #AI safety #privacy leaks

Official accounttheverge@Emma RothSource

Oct 16

Chinese content联邦微调中的隐私泄露与后门投毒研究：FedAvg 攻击与 FLTrust 防御实测

arXiv 上的联邦学习攻防实验报告，服务端能从一次更新里偷出标签集，30% 恶意客户端就能植入后门，FLTrust 能防住，代码可直接跑。

#FedAvg #FLTrust #federated learning #data poisoning

arXiv cs.LG@Vasileios Perifanis, Nikolaos PavlidisSource

Sep 21

Papers BuildersSelected09:50

CIPL: A Channel-Aware Framework for Recoverable Privacy Leakage in LLM Agents

This academic paper introduces a new framework for evaluating privacy risks in LLM agents, which is important for anyone developing or using these systems to understand potential data exposure.

#CIPL #agents #privacy leaks #model evaluation

arXiv cs.AI@Tao Huang et al. (8)Source

Sep 11

Papers BuildersSelected11:22

Predicting Privacy Leakage from Weight Spectral Density

This research provides a new method for evaluating privacy risks in machine learning models, which is important for developers and researchers working on AI systems.

#Membership Inference Attacks #privacy leaks #membership inference

arXiv cs.LG@Richard J. Preen, Jim SmithSource

Sep 1

PapersSelected09:20

Boundary Metadata Collapse in Multi-Agent LLM Handoffs

Developers using multi-agent systems should implement explicit boundary constraints to reduce privacy leakage by over 50% compared to vague language.

#multi-agent systems #LLM #GPT-5 #DeepSeek-R1-32B

arXiv: DeepSeek@Yian Wang et al. (4)Source

Aug 24

Chinese contentPUN协议：用于评估以人为中心的LLM的未知姓名

想了解如何评估LLM在隐私和偏见方面的表现？这篇论文提出了一个新方法，值得一读。

#LLM evaluation #privacy leaks #bias detection

arXiv cs.AI@Dimitri Staufer et al. (3)Source

Jul 29

IndustryMulti-source72°07:14

Chinese content数百条 Claude 聊天记录被谷歌索引，Anthropic 回应称用户主动分享所致

Claude共享链接被谷歌索引，明文暴露简历、医疗数据甚至API密钥，赶紧去设置里检查并撤销已分享的对话。

#Claude #Anthropic #Google #AI safety

IT之家Source

Jul 28

Industry72°04:26

Chinese content你的Claude共享聊天和Artifacts可能已被Google索引

你的Claude共享内容可能被Google搜到，赶紧检查哪些对话是公开的！

#Claude #Artifacts #Google #AI safety

techcrunch@Lorenzo Franceschi-BicchieraiSource

Jul 17

IndustryMulti-source15:30

Chinese content新 AI 漏洞披露：可从 Claude 记忆中泄露姓名与工作单位等信息

安全研究员发现 Claude 记忆功能漏洞，能骗它把姓名公司发到外部网站。用 Claude 时小心点，别让它存敏感信息。

#Claude #Anthropic #AI safety #vulnerabilities

IT之家Source

Jun 13

Chinese content国家安全部发布使用“AI中转站”风险提示：警惕个人数据泄露

官方警告AI中转站数据裸奔风险

#data security #privacy leaks

IT之家Source