Falcon Exposure Management Now Available for Third-Party EnvironmentsJun 16, 2026
Falcon Exposure Management Now Available for Third-Party EnvironmentsJun 16, 2026
2026-06-21 归档
今日共收录 748 条安全情报，包含 44 个 CVE，388 篇安全通告，以及 64 篇研究论文。
← 返回归档列表After Executive Order 14409: Next Steps for Securing AIJun 17, 2026
After Executive Order 14409: Next Steps for Securing AIJun 17, 2026
Cyber Defense Summit 2026Register today to reserve your spot
Cyber Defense Summit 2026Register today to reserve your spot
Cyber Defense Summit 2026Crafted to equip elite security professionals with the strategies, tools, and insights needed to outmaneuver increasingly sophisticated, AI-enabled adversaries and build resilient cyber ecosystems.Register now
Cyber Defense Summit 2026Crafted to equip elite security professionals with the strategies, tools, and insights needed t
Security TalksJoin our security experts in this ongoing series as they explore the latest AI innovations across our security product portfolio, threat intelligence best practices, and more.Watch on-demand
Security TalksJoin our security experts in this ongoing series as they explore the latest AI innovations across our secu
New Abuse of the ClickOnce Technology, Part 2: Stop Threat Actors from Clicking Once and Staying Forever
New Abuse of the ClickOnce Technology, Part 2: Stop Threat Actors from Clicking Once and Staying Forever
New Abuse of the ClickOnce Technology, Part 1: The Inner Workings of ClickOnce Application Deployment
New Abuse of the ClickOnce Technology, Part 1: The Inner Workings of ClickOnce Application Deployment
Threat Brief: Mitigating Large-Scale Credential Attacks
We provide guidance for preparing for and mitigating large-scale credential attacks, focusing on recent campaigns target
Klue OAuth breach victim list grows as Icarus hackers claim attack
Market intelligence platform Klue has publicly confirmed a recent security incident that allowed threat actors to steal
Hackers exploit info disclosure bug in Gravity SMTP WordPress plugin
Threat actors are exploiting an unauthenticated information disclosure vulnerability in the WordPress plugin Gravity SMT
The Gentlemen RaaS Uses GentleKiller EDR Framework Targeting 400 Security Processes
The Gentlemen ransomware-as-a-service (RaaS) operation is actively developing and maintaining a suite of endpoint detect
New Prinz Eugen ransomware prioritizes recent files for encryption
A new ransomware operation named 'Prinz Eugen' prioritizes recently modified files for encryption and leaves no ransom n
Multi-View Decompilation for LLM-Based Malware Classification
该论文针对基于大语言模型（LLM）的恶意软件分类中反编译器视图单一的问题展开研究。现有方法通常仅依赖一种反编译器（如Ghidra）生成的伪C代码，但反编译器是有损的启发式工具，不同反编译器可能揭示同一二进制文件的不同侧面。作者构建了一个包含
Calibration Without Comprehension: Diagnosing the Limits of Fine-Tuning LLMs for Vulnerability Detection in Systems Software
该论文研究了大语言模型（LLM）在系统软件漏洞检测中的真实能力，质疑模型在基准测试中的高分是否源于真正的安全推理还是数据污染。作者提出了CWE-Trace框架，一个基于834个手动整理的Linux内核样本构建的漏洞检测数据集，涵盖74个CW
LLM agent safety, multi-turn red-teaming, jailbreak benchmarks, adversarial robustness, safety-critical systems
本文提出了 NRT-Bench，一个针对大型语言模型（LLM）代理在多轮对抗性攻击下的鲁棒性评估基准。研究背景是LLM代理越来越多地被提议用于安全关键系统的监督组件，但它们在持续的、自适应的对抗压力下的鲁棒性尚未被充分表征。作者将基准实例化
Quantization as a Malicious Task: Removing Quantization-Conditioned Backdoors via Task Arithmetic
模型量化作为一种广泛采用的技术，能够在资源受限设备上部署深度神经网络时减少内存使用和推理成本。然而，最近的研究揭示了一种新的安全威胁——量化条件后门（QCBs），即模型在全精度下表现正常，但仅在量化后激活恶意行为。现有的防御方法通常修改量化
FFinRED: An Expert-Guided Benchmark Generation and Evaluation Framework for Financial LLM Red-Teaming
FinRED是一个面向金融领域大语言模型（LLM）安全评估的专家引导红队框架，由韩国金融安全研究院（FSI）联合学术界开发。现有安全基准主要针对通用对抗场景，忽略了金融特有的风险，如监管合规违规、欺诈辅助和系统性信任侵蚀。FinRED首先构
SafeSpec: Fast and Safe LLM via Dynamic Reflective Sampling
大型语言模型（LLM）的推测性推理（speculative inference）虽然能加速解码过程，但本身不提供任何安全保障。现有的安全防御方法大多与推测性推理不兼容：它们要么引入额外计算，要么破坏草稿-验证机制，从而抵消加速优势。这揭示了
Analyzing the Narration Gap in LLM-Solver Loops
该论文关注于将形式化工具（如SAT和SMT求解器）集成到语言模型推理流程中所产生的“叙述差距”问题。在安全或安全关键场景中，问题可被形式化为逻辑公式，求解器提供可验证的正确答案，但最终用户看到的是由LLM将求解器输出转化为自然语言叙述的结果
FloatDoor: Platform-Triggered Backdoors in LLMs
大型语言模型（LLM）日益部署在软件工程等敏感场景中，其输出直接影响下游工件。近期研究发现，同一模型在不同部署平台上可能产生可测量的输出差异，这是由非结合浮点运算和不同的内核实现所致。本文研究了这种平台依赖可变性的安全影响，揭示了LLM部署