Cyber Security Daily Radar

2026-10-04 归档

今日共收录 931 条安全情报，包含 128 个 CVE，457 篇安全通告，以及 95 篇研究论文。

← 返回归档列表
安全公告2026-10-03 发布INFO

In Bouncy Castle for Java before 1.86, the Messaging Layer Security (MLS, RFC 9420)...

CVE-2026-71885

In Bouncy Castle for Java before 1.86, the Messaging Layer Security (MLS, RFC 9420) implementation did not bind an X.509

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-287#severity:critical
安全公告2026-10-03 发布INFO

In Bouncy Castle for Java before 1.86, BLS12_381BasicScheme.keyValidate, and so...

CVE-2026-71891

In Bouncy Castle for Java before 1.86, BLS12_381BasicScheme.keyValidate, and so BLSPublicKeyParameters and every BasicSc

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-347#severity:high
安全公告2026-10-03 发布INFO

In Bouncy Castle for Java before 1.86, the high-level OpenPGP API accepted a data signature made...

CVE-2026-71887

In Bouncy Castle for Java before 1.86, the high-level OpenPGP API accepted a data signature made by a signing subkey who

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-345#severity:high
安全公告2026-10-03 发布INFO

In Bouncy Castle for Java before 1.86, the high-level OpenPGP certificate API accepted a third...

CVE-2026-71886

In Bouncy Castle for Java before 1.86, the high-level OpenPGP certificate API accepted a third-party certification or tr

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-285#severity:high
安全公告2026-10-03 发布INFO

The VikAppointments Services Booking Calendar plugin for WordPress is vulnerable to arbitrary...

CVE-2026-87115

The VikAppointments Services Booking Calendar plugin for WordPress is vulnerable to arbitrary file deletion due to insuf

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-22#severity:critical
安全公告2026-10-03 发布INFO

The WPMobile.App – Android and iOS App Builder plugin for WordPress is vulnerable to Stored Cross...

CVE-2026-103421

The WPMobile.App – Android and iOS App Builder plugin for WordPress is vulnerable to Stored Cross-Site Scripting via the

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-79#severity:medium
安全公告2026-10-03 发布INFO

The Photo Gallery by 10Web – Mobile-Friendly Image Gallery plugin for WordPress is vulnerable to...

CVE-2026-92974

The Photo Gallery by 10Web – Mobile-Friendly Image Gallery plugin for WordPress is vulnerable to Reflected Cross-Site Sc

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-79#severity:medium
安全公告2026-10-03 发布INFO

The Kubio AI Page Builder WordPress plugin before 2.9.3 does not limit its widening of the...

CVE-2026-88783

The Kubio AI Page Builder WordPress plugin before 2.9.3 does not limit its widening of the allowed HTML elements to the

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-79#severity:high
安全公告2026-10-03 发布INFO

The TillKit WordPress plugin before 1.0.5 does not require the hard-coded, publicly known PIN of...

CVE-2026-91078

The TillKit WordPress plugin before 1.0.5 does not require the hard-coded, publicly known PIN of the privileged POS acco

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-287#severity:high
安全公告2026-10-03 发布INFO

The Kubio AI Page Builder WordPress plugin before 2.9.3 does not validate the URI scheme of a...

CVE-2026-88782

The Kubio AI Page Builder WordPress plugin before 2.9.3 does not validate the URI scheme of a user-supplied value before

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-79#severity:medium
安全公告2026-10-03 发布INFO

The Groundhogg — CRM, Newsletters, and Marketing Automation plugin for WordPress is vulnerable to...

CVE-2026-97644

The Groundhogg — CRM, Newsletters, and Marketing Automation plugin for WordPress is vulnerable to Privilege Escalation v

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-269#severity:high
安全公告2026-10-03 发布INFO

The The All in One SEO – AI SEO Plugin to Boost SEO Rankings & Traffic (Schema, Local SEO,...

CVE-2026-100152

The The All in One SEO – AI SEO Plugin to Boost SEO Rankings & Traffic (Schema, Local SEO, Sitemap & SEO Insights) plugi

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#cwe-94#severity:medium
安全公告2026-10-02 发布INFO

gitea-runner: workflow container.options passes host namespaces and capability flags to job container when privileged mode is disabled

CVE-2026-73802

### Summary act_runner appends workflow-controlled `jobs..container.options` directly to the Docker HostConfig for the

历史摘要 · 未重新核验 · GitHub Advisory

#advisory#go#cwe-269#severity:critical
学术论文2026-10-01 发布INFO

MOMAT: Mixture of Multiple Atlases for Low-Power Jailbreak Defense of Quantized LLMs

该论文关注量化大语言模型（qLLM）在边缘设备上的安全风险。量化虽带来低延迟、低能耗的部署优势，但会削弱模型的对齐防护能力，使 qLLM 更易被越狱提示绕过。作者提出 MOMAT（Mixture of Multiple Atlases），一

历史摘要 · 未重新核验 · arXiv cs.CR

#cs.AI#llm-jailbreak-defense#quantized-llm#edge-ai-security#rag-safety#compute-in-memory
学术论文2026-10-01 发布MEDIUM

Walking the Embedding Space: Datastore Extraction from Multimodal RAG

该论文关注多模态检索增强生成（MRAG）引入的新型安全风险。MRAG 通过将外部、最新的多模态知识注入多模态大模型（MLLM）的生成过程，有效降低幻觉、提升事实性，并具备成本优势，但同时把原本封闭的私有数据存储（datastore）暴露为新

历史摘要 · 未重新核验 · arXiv cs.CR

#cs.AI#rag-security#multimodal#data-extraction#embedding-space#llm
学术论文2026-09-30 发布MEDIUM

From A2A Attacks to Envelope-Layer Defense: Red-Teaming Evaluation of LLM Agents and a Three-Layer Isomorphic Attack-Defense Model

随着 ACP、A2A 等 Agent 交互协议的普及，基于大模型的智能体正从单体工具调用走向多智能体协作，由此引入了新的安全威胁：在 A2A 场景下，来自远端 peer 的任务会被本地 Agent 默认视为合法请求，这为间接提示注入提供了天

历史摘要 · 未重新核验 · arXiv cs.CR

#agent-security#llm#prompt-injection#a2a#multi-agent#red-teaming
学术论文2026-10-01 发布LOW

OverAct: Measuring and Mitigating Proactive Over-Authorization in LLM Tool-Calling Agents

该论文研究 LLM 工具调用型智能体（tool-calling agents）在结构化工具调用场景中的一种隐私风险行为，作者将其命名为“主动式越权（proactive over-authorization）”，即智能体在完成用户请求时，主动

历史摘要 · 未重新核验 · arXiv cs.CR

#cs.CL#agent-security#llm#tool-calling#privacy#over-authorization
学术论文2026-09-30 发布INFO

Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection

该论文研究的是参数高效微调（PEFT）场景下的后门防御问题。随着大语言模型（LLM）与 LoRA 等轻量微调方法的广泛采用，攻击者可以通过污染微调数据或发布被植入后门的适配器（adapter），使模型在特定触发条件下产生恶意输出，而基座模型

历史摘要 · 未重新核验 · arXiv cs.CR

#cs.AI#cs.LG#llm#backdoor#lora#peft
学术论文2026-09-30 发布INFO

Backdoor Containment via Expert Quarantine and Shutdown in LLMs

该论文研究大型语言模型（LLM）后门防御问题。被植入后门的 LLM 在正常输入上表现正常，但在隐藏触发器下会产生攻击者指定的输出。现有防御覆盖训练前、训练中、训练后和推理时四个阶段，底层策略可归为两类：一是抑制后门学习，例如过滤投毒数据或在

历史摘要 · 未重新核验 · arXiv cs.CR

#cs.AI#cs.LG#llm#backdoor-defense#moe#model-security
学术论文2026-10-01 发布MEDIUM

Sleeping Secrets: How Fine-Tuning Reawakens Privacy Risks in Language Models

该论文研究的是大语言模型（LLM）微调过程中的隐私泄漏问题。此前的多项工作已经表明，微调不仅能让模型适配下游任务，还会“唤醒”那些通过直接查询已经无法获取的私有信息——即所谓的微调恢复攻击（fine-tuning recovery atta

历史摘要 · 未重新核验 · arXiv cs.CR

#llm#privacy-leakage#fine-tuning#lora#data-free-attack#model-alignment