Beyond the Model: Harnessing Frontier AI for Stronger Cyber Defense
Beyond the Model: Harnessing Frontier AI for Stronger Cyber Defense
2026-08-31 归档
今日共收录 884 条安全情报，包含 132 个 CVE，408 篇安全通告，以及 86 篇研究论文。
← 返回归档列表make smarter security investments and mitigate future risks
make smarter security investments and mitigate future risks
Inside Astaroth's New Spambot Component
Inside Astaroth's New Spambot Component
Falcon Platform IOAs Arrive in Falcon Next-Gen SIEM to Identify New Threats
Falcon Platform IOAs Arrive in Falcon Next-Gen SIEM to Identify New Threats
CrowdStrike 2026 Threat Hunting Report: Exploitation Window Closes as AI Use Accelerates
CrowdStrike 2026 Threat Hunting Report: Exploitation Window Closes as AI Use Accelerates
CrowdStrike Threat Hunts for Shell Command Obfuscation on VMware ESX
CrowdStrike Threat Hunts for Shell Command Obfuscation on VMware ESX
Google is a Leader in the IDC MarketScape: Worldwide Cybersecurity Consulting Services 2024 Vendor Assessment
Google is a Leader in the IDC MarketScape: Worldwide Cybersecurity Consulting Services 2024 Vendor Assessment
August 2026 Patch Tuesday: One Exploited Zero-Day and 62 Critical Vulnerabilities Among 415 CVEsAug 11, 2026
August 2026 Patch Tuesday: One Exploited Zero-Day and 62 Critical Vulnerabilities Among 415 CVEsAug 11, 2026
Teaching AI to Reason Through Detection Triage
Teaching AI to Reason Through Detection Triage
Benchmaxxing: When the Benchmark Becomes the TargetAug 19, 2026
Benchmaxxing: When the Benchmark Becomes the TargetAug 19, 2026
Anthropic is cutting Claude Code's current weekly limits by 17%
Anthropic is permanently increasing Claude Code's standard weekly usage limits by 25% for Pro, Max, Team, and seat-based
FulcrumSec claims Manchester Airports hack, theft of 86 GB of data
FulcrumSec claims it stole 86 GB of data from Manchester Airports Group. BleepingComputer validated one traveller's reco
Anthropic warns infostealer malware is hijacking Claude sessions to drain usage
Anthropic is warning some Claude users that infostealer malware on their PCs has stolen active Claude login sessions, al
The Impact of Magma: A Ground-Truth Fuzzing Benchmark
Magma 是一个开源、基于真实漏洞（ground-truth）的模糊测试基准测试工具集，旨在解决模糊测试领域长期缺乏统一、可重复评估标准的问题。该工具最初随一篇发表于 ACM SIGMETRICS 2021 的研究论文发布，而本文则是一篇
Compared to What? A Human-Anchored Security Benchmark for LLM-Generated Infrastructure-as-Code
这篇论文提出并实现了一个面向LLM生成的基础设施即代码（IaC）的安全基准评测框架，核心关注点是：当大模型被用来编写IaC时，其生成结果是否比人类工程师更容易引入安全缺陷。作者首先指出现有评估方法仅报告模型生成IaC的原始漏洞数量，但缺乏与
LAAF: A Layered Accountability Architecture Framework for LLM Applications
该论文针对大型语言模型（LLM）在医疗、司法、金融和公共服务等高风险领域应用中出现的责任归属问题，提出了一种分层责任架构框架（LAAF）。研究背景是：LLM 在这些场景中即使输出缺乏依据或内容错误，也常被当作权威信息，一旦造成损害，难以明确
Recognition Without Enforcement: Configuration-Dependent Failures in LLM Agent Instruction Arbitration and External Control
这篇论文研究了 LLM 智能体在系统提示、用户输入、记忆和工具等来源之间进行指令仲裁时的一个关键缺陷，作者称之为“识别-执行鸿沟”（recognition-enforcement gap）。他们发现，虽然模型可以从激活值中线性解码来源格式特
LongPIBench: A Long-Context Benchmark for Prompt Injection
该论文针对大语言模型在真实应用中面临的提示注入攻击安全风险，指出现有提示注入基准大多局限于短上下文输入，导致长上下文场景下的攻击与防御研究严重缺失，并造成对当前防御效果的高估。为弥补这一空白，作者提出 LongPIBench，一个面向长上下
When Verified Source Becomes Attack Input: Defending Smart Contracts Against LLM-Based Vulnerability Scanning
本文研究智能合约源代码公开披露机制在LLM时代面临的新威胁。智能合约通常需要在区块链浏览器上公开源代码并与其字节码进行验证，以建立用户信任，但这使得攻击者可以利用LLM代理对这些公开代码进行大规模漏洞扫描和利用。作者提出DeLLMGuard
Layered LLM Defenses as an Ensemble: Access Tiers, Inference Cost, and the Measured Failure Correlation Between Defense Layers
本文研究大型语言模型（LLM）多层防御堆叠的效果。防御者通常通过叠加多个防御层来增强安全性，但堆叠本质上是集成学习，其有效前提是各防御层对不同输入产生失败，即失败相关性低。然而现有 LLM 安全文献推荐这种堆叠做法，却从未测量这一前提。作者