⚡Megatron LLM Hub
全部
Twitter

推特安全流

头条8

开源工具可直接扫描自托管 Agent、MCP 服务与基础设施,AI 红队上手成本低

AI-Infra-Guard v4.6.6 — A full-stack AI Red Teaming platform securing AI ecosystems via Agent Scan, Skills Scan, MCP scan, AI Infra scan and LLM jailbreak evaluation. https://t.co/jDbZcLnbwz https://t.co/jVBnHuSPvi

ai_agentllm工具红队 @KitPloit 原文 ↗

路径穿越写文件即可利用且无需交互,对外暴露的 SfB 服务器应尽快打补丁

RT @ArmadinSecurity: Armadin researcher @mhskai2017 has uncovered an authenticated RCE vulnerability in Skype for Business Server (CVE-2026-70297, CVSS 8.8). An attacker can exploit a path traversal vulnerability to write arbitrary files to the server while holding a valid token that can access any chat room with file-upload feature supported. No user interaction required. Patch immediately. Updates are available for Skype for Business Server 2019 CU8, 2015 CU13, and Subscription Edition CU1. Microsoft advisory: https://t.co/AtDO2Tkg9L

rcecve路径穿越企业服务 @0xocdsec 原文 ↗

利用条件明确且 curl 部署极广,8.23.0 及多个分支已修复,需尽快升级。

RT @JFrogSecurity: 🚨 RCE in curl (CVE-2026-92392, HIGH Severity) 🚨 A malicious HTTP/2 server can turn a use-after-free during uploads into remote code execution. Prerequisites: ● curl -T <file> over HTTP/2 to an attacker-controlled server ● <file> is ~135KB+ and contains <word> <word> <any>\n\n anywhere ● -Z / --parallel for exploitation when PIE/ASLR is on Affected: 8.12.0–8.22.0 Fix: 8.23.0 (or 8.20.2 / 8.16.2 / 8.14.3)

rcecve供应链 @0xocdsec 原文 ↗

单次补丁量创纪录,317 个 RCE、677 个提权,在野 0day 需优先处置。

RT @BleepinComputer: 🚨 BREAKING: Microsoft's October 2026 Patch Tuesday is MASSIVE! Microsoft has fixed a record-breaking 1,411 security vulnerabilities, including 70 critical flaws and 3 zero-days, one of which is actively exploited in attacks. 🔥 317 remote code execution flaws ⚠️ 677 privilege escalation flaws 🛡️ 40 security feature bypass flaws 💻 Time to get patching!

cvercelpe0day @NetAskari 原文 ↗

覆盖 2018–2026 野外样本,可直接用于检测规则与指纹匹配

RT @YunZhengHu: Fox-IT’s full Cobalt Strike Beacon Corpus now available as an open dataset. 56,843 raw beacons + parsed configs from 190,182 sightings across 30,604 Team Servers, collected 2018–2026. https://t.co/S9YzJ35NaP

c2数据集威胁情报工具 @_RastaMouse 原文 ↗

简化会话劫持后的 Cookie 筛选,绕过 MFA 流程更快

🚨 EVILGINX EXTENSION IS OUT🚨 I am happy to say the free Evilginx Chrome extension has been approved and is finally available for download. 🥳 You can now safely import your captured session cookies, knowing that the cookie manager you're using came from a reputable source, along with the "trust me bro" quality seal. For a long time, I wanted to make the task of finding the cookies that hold session tokens much easier. With the Evilginx extension, you can now speed up the trial-and-error process when testing multiple cookie configurations to see whether you're still authenticated after a page reload (no more restoring cookie backups from external JSON files). I made this extension for myself, and I use it to kickstart the development of every single phishlet I make. Now you can too! 💪 To properly celebrate the release and help you get the most out of the extension, I released a walkthrough video with a rundown of all the features and a hands-on tutorial on how to use it to kickstart the development of your own phishlet for GitHub. Enjoy and happy phishing! 💗🪝🐟 Disclaimer: AI was used to develop the extension. UX, UI design, and minor code tweaks originated from a human brain. Code was read using human eyes. P.S. GitHub repo will arrive later. 🔗 Links can be found below 👇

工具浏览器会话劫持mfa @mrgretzky 原文 ↗
必看21

无锁文件时仅 Go/vcpkg 全量校验,供应链风险被量化

RT @trailofbits: We tested 25 package managers against a simulated malicious repository. Without a lockfile, only Go and vcpkg verified every download in our tests. With a lockfile, it's fewer than half. https://t.co/bFh0NIfhmG

供应链报告工具 @0xocdsec 原文 ↗

来自官方商店,绕过信任边界,企业账号面临 MFA 劫持

RT @wbmmfq: Analysts @HuntressLabs have been seeing an uptick of malicious MFA apps from the Microsoft Store the last few days. In some cases, they're being delivered via tech support scams through Teams, in others via malvertizing. We've been digging into it, and hopefully we'll have more info soon.

事件钓鱼供应链 @0xocdsec 原文 ↗

自托管虚拟化隔离边界被突破,复现思路与工具链已公开,补丁前需评估暴露面。

RT @0x4D31: now that the KVM escape details are out (https://t.co/5QFoYyonbX)… if you’re wondering how i managed to rediscover it in under 2 hours after seeing the vercel bounty screenshot: i gave openai’s gpt-5.6-cyber (daybreak red) access to a devbox to experiment and explore different directions for 1.5 hours. then i spotted that the word after “guest-controlled” in the screenshot wasn’t properly redacted. those 3 dots gave away “virtio”. might seem like a small hint, but i passed it to my agent, and it got excited and found the vulnerability in 30 minutes. :)) opsec lesson right there.

逃逸kvmai_agentpoc @0xocdsec 原文 ↗

不含内存破坏即可全链控制浏览器,说明逻辑缺陷的实战利用价值被低估。

RT @d3vc0r3: Orange Tsai (@orange_8361) is taking the #HEXACON2026 stage on Oct 16 in Paris with the logic-only Microsoft Edge full chain he used at #Pwn2Own Berlin 2026.🍊🏆 Five logic bugs, chained. No renderer bugs, no memory corruption, no AI, no collisions. https://t.co/ubYP4gQV2g

浏览器绕过议题poc @0xocdsec 原文 ↗

Linux 网络栈缺陷导致跨 hypervisor VM 逃逸,CVSS 9.3,多租户与云环境影响面极大

RT @PaulosYibelo: The details of the long awaited KVM Escape are published. PwnEscape leads to VM escapes across multiple hypervisors due to a super cool Linux networking bug. CVE-2026-99999 is a 9.3 CVSS critical rated KVM escape ⛓️ https://t.co/CXLvvjLxyF

逃逸cve内核云 @0xocdsec 原文 ↗

Agent 经 MCP 调用逆向能力的现成范例,可用于评估插件与权限设计风险

RT @tonbistudio: Hermes Agent now has an official plugin for Reverse Engineer Anything by @morluto ! I made a short video breaking down the plugin and showing how it works hermes plugins install rea That one command will get you set up with the proper MCP config and a special skill your agent can use!

ai_agentmcp工具逆向 @morluto 原文 ↗

主流虚拟化平台受影响,可从不可信 VM 直达宿主内核读写与 root 执行。

RT @pwn_ai: Introducing PwnEscape 🦉: a full KVM guest-to-host escape via a Linux kernel networking flaw affecting industry standard hypervisors. CVE-2026-99999 went from untrusted VM code to host-kernel memory reads, controlled writes, and root command execution. More below: https://t.co/KrNtI2zzCu

逃逸内核cve云 @moyix 原文 ↗

同机低权进程即可接管本地 AI 客户端账号并读取受保护文件,细节已公开。

Neat ChatGPT bug 🐛 Found manually (even after @OpenAI “pointed Astra at its own systems until it ran out of bugs") Local (unpriv'd) attacker: • Hijack your Dot • Take over your account • Access protected macOS files +more 👀 @OpenAI's $500 bounty donated to @objective_see https://t.co/SUHDTYWJts

ai_agentllmlpemacos @patrickwardle 原文 ↗

广告管道本身即感染途径,屏蔽广告是目前主要缓解手段

Again, fantastic work by @billmarczak @jsrailton and all the other folks at @citizenlab : This report exposes zero-click spyware delivered via targeted ads, which we call AdFection, and is the first technical demonstration that mercenary spyware is being distributed via targeted advertisements."https://t.co/WePS3AmsnK

恶意广告间谍软件报告事件 @NetAskari 原文 ↗

密钥源自官方文档默认值,同类暴露配置可自查并检测

571 IIS servers hijacked using machine keys from public docs. @soolidsnakee and @k33b0i break down REF3927 at @virusbtn VB2026 • Today, 15:00 CEST in the Green room. Talk abstract and kill chain outline: https://t.co/HqODWpz28b Full blog: https://t.co/3XYkOKR4J5 https://t.co/aviLgTheMj

iisrce默认凭据报告 @elasticseclabs 原文 ↗

自托管 Agent 在逆向场景的工程化取舍,可直接迁移到自有 agent 设计。

RT @r00tth3w0r1d: https://t.co/zRu0Y4G57Z + real Touhou reconstruction with @TouhouLab changed how I think about agentic RE: • autonomy &gt; rigid pipelines • repo/Git &gt; chat memory • evidence &gt; confidence; “unknown” is valid TH08 is where a lot of this got battle-tested: https://t.co/S7qBquHcBK

ai_agent工具报告 @morluto 原文 ↗

沙箱应用无需特殊权限即可触发,两平台 PoC 已公开。

RT @_MrNiko: 🚨 CVE-2026-43748 ANE kernel OOB write direct-path user client. no special entitlements. sandboxed app on iOS and macOS. kANEMaxBuffers allows 0xff surfaces. memoryMapRequest only has room for 0x80 entries of 0x10. the capacity check runs after the descriptor loop already wrote past the table. up to ~0x7f0 bytes into data.kalloc.3072. PoCs for both platforms: https://t.co/Xp4OvAI9yM credit: vschko #iOS #ExploitDev #InfoSec

pociosmacos内核 @kernullist 原文 ↗

CVSS 8.8,PoC 由原研究者发布,完整接管仍需沙箱逃逸。

RT @ThreatWire_: 🚨 PoC RELEASED — Chrome CVE-2026-91737 PDF use-after-free (High). A crafted web page can run code inside Chrome's renderer sandbox. Published by the credited researcher, SeungMyung Lee (@sm1ee), 4 weeks after the fix. ThreatWire has not run or verified it. ✅ Fixed in Chrome 153.0.8010.47 (Sep 15) NVD: CVSS 8.8 (CISA ADP) 📌 Sandbox only: full compromise would need a separate sandbox escape. No chain has been claimed. • Not in CISA KEV • No known exploitation Check chrome://settings/help and restart the browser to apply the update. Full breakdown 👇 https://t.co/xXqnTG1Hj3 #Chrome #CVE #cybersecurity

浏览器poccve @0xocdsec 原文 ↗
推荐15

120B 稀疏模型的本地速度实测,自托管 agent 选型参考

RT @Sentdex: Nemotron 3.5 Super VL (120B A12B) is now the leading US open source model. Congrats to the NVIDIA AI team! I am still doing testing of this model, but I think the ranking would be even stronger if we were just measuring agentic coding performance. This one may very well be a solid choice for a speedy local coding agent model I'm running many things in parallel, will get proper average and max numbers in time, but just randomly prodding this model for some quick speed checks and over 550 tok/sec. All vanilla settings. Once optimized 600+, maybe even 650+, will be trivial, and so far it does appear to be a token efficient problem solver. Really exciting to see what looks like a solid release so far from nvidia!

llmai_agent本地部署 @0xocdsec 原文 ↗

自建 Agent 的审计、权限边界与行为基线设计可直接借鉴该评估思路。

Task completion is an incomplete measure of AI work. Agent traces let us infer an organization’s norms and intents, then evaluate behavior against its purpose, practices, and policies. Securing AI work requires modeling the organization. https://t.co/8LC18xSJw0

ai_agentllm报告 @matterpreter 原文 ↗

安全设备自身被攻陷的根因分析,与近日同类网关事件可相互印证。

RT @Enno_Insinuator: Looks quite similar to what we did with FireEye appliances a while ago, and probably same root causes. 'Playing With Fire: Attacking the FireEye MPS' from @_fel1x https://t.co/fdU9YEPfmV [PDF] https://t.co/Qp7qPuAJKo

固件提权报告poc @0xocdsec 原文 ↗

可自托管的多模型 Agent 做二进制分析,含沙箱执行便于本地部署。

RT @oneandonlyhusam: Introducing BINNY, now available in public beta. An open-source, agentic AI harness designed for reverse engineers and security researchers. BINNY combines multi-LLM support, Ghidra integration, sandboxed execution, and autonomous binary analysis to streamline reverse engineering workflows. Bring your own model. Spend less time on repetitive tasks and more time on research. https://t.co/aFrzPdSHwV #ReverseEngineering #CyberSecurity #OpenSource

ai_agentllm工具 @0xpwnie 原文 ↗

给出 AI 蠕虫威胁的量化反驳,避免被厂商叙事带偏。

RT @BlaiseBits: I went through the anthropic "IOT AI-Worm" blog post last night. This "sign of life" was 15 android TV modules running Gemma4-26B-A4B. This absolute powerhouse of inference ran at a earth shattering 0.38 tps.... They later extrapolate to what a AI Worm Swarm the size of the Kimwolf could generate inferance of roughly 80k tps. All that sounds oh so scary till you realize: 1) It's slow af 2) any node going offline kills the "worm agent" 3) network latency is a bizzich Possible, but not probable. This would be the world's SLOWEST replication worm as well as being hella easy to identify with, get this, cyber security controls we've had for decades. The report even estimates ~3 days for a replication spread on these IOT devices, meaning any node (Android TV Box) going offline BTFOs that attempt, or if it's rebooted that inference instance is dunzzos without persistence.... Which is likely gonna become a problem if a user can't even use their device because it's too busy with matrix multiplication and gets a hard reset, perminanlty removing it from the swarm. Anthropic did hedge some these thoughts in the blog, but that part is skimpy and burried by the FUD/Doomer crowd. https://t.co/0Gm63PgvP6

ai_agentllm报告 @mrgretzky 原文 ↗

Agent 可直接驱动载荷生成与免杀,本地 MCP 部署的攻击面随之扩大。

We are preparing an optional MCP server for the next version of MacroPack, ShellcodePack, and DarwinOps! This interface will come in addition to the existing CLI and GUI for our customers who want to integrate redteam tooling with agentic AI! 🤖 #redteam https://t.co/Jtvgqp90bS

ai_agentmcp红队工具 @BallisKit 原文 ↗

免注入完成 hook 与远程内存读写,可绕过常见基于注入的检测。

Cross-process instrumentation for Windows x64 in TypeScript ✅ Hook Functions ✅ Redirect Threads, and ✅ Read/write memory in another process, with no DLL injection and no CreateRemoteThread. https://t.co/WpEB1n1g5M

工具windows注入绕过 @ipurple 原文 ↗

借受害者凭据转发请求,可绕过出站与网络边界限制

RT @Teach2Breach: https://t.co/zKV2rbGXx9 update: new technique, BM-T3009 "Browser Relay" uses an attacker browser extension with websockets to turn compromised browser into a relay, including making requests with credentials. Learn how to build it for your red and purple team engagements. https://t.co/hMaPICTKPX

浏览器c2绕过工具 @0xpwnie 原文 ↗

指标聚焦认证与 SSO 仿冒域名,可直接导入狩猎查询

🔑 Fresh Microsoft Defender indicator update for 𝗦𝘁𝗼𝗿𝗺-𝟯𝟮𝟬𝟰, focused on 𝗽𝗮𝘀𝘀𝗸𝗲𝘆/𝗠𝗙𝗔-𝘁𝗵𝗲𝗺𝗲𝗱 𝗱𝗼𝗺𝗮𝗶𝗻𝘀 that strongly resemble authentication, enrollment, SSO, or support workflows.🫡 ​ #Threathunting https://t.co/cNTlduXkvV

钓鱼mfaioc威胁情报 @0x534c 原文 ↗

一键测试多种访问控制绕过,成功后直接给出复现命令

unKover — Bash 403/401 access bypass tester that probes a URL with header spoofing, method tampering, path normalization, and encoding techniques, returning a curl PoC on success. https://t.co/B68X8lRDCE https://t.co/umsxtA4vAg

工具绕过poc @KitPloit 原文 ↗

间接系统调用加 ETW/AMSI 补丁,服务账户即可提至 SYSTEM。

CouchPotato — Windows privilege escalation tool that patches ETW and AMSI and uses indirect syscalls to abuse SeImpersonatePrivilege, escalating service accounts or admins to NT SYSTEM. https://t.co/V3jY3o08CX https://t.co/FjgziR6Vxi

提权工具绕过windows @KitPloit 原文 ↗

同一漏洞的第二方复现与披露时间线,可交叉验证触发条件。

RT @destr4ctt: Sharing details &amp; PoC for CVE-2026-43748, a heap OOB-write in ANEHWDevice::ANE_ProgramCheckandPrewireBuffers_gated (Apple Neural Engine) Vulnerability was fixed in iOS 26.6 update. And for some reason, Apple missed me in advisory... Frustrating :( https://t.co/HAmApCS4EA

pociosmacos @0xocdsec 原文 ↗
更多72

线程池回调执行是常见免杀手法,附完整代码与引用

Shellcode Execution via CreateThreadpoolWait New Medium post. Today we are looking at a shellcode execution technique that abuses the Windows thread pool (CreateThreadpoolWait) to dispatch shellcode on a worker thread https://t.co/V7O0gNDQ5m Credits: https://t.co/vzCUcZhRU0 https://t.co/loWWBF57lD

shellcodewindowspoc @Salsa12__ 原文 ↗

内核利用入门资料稀缺,含贴近真实场景的简化案例

RT @andersonc0d3: We finally got this series out. We tried to simplify Linux kernel exploitation for those interested in getting started. There are plenty of resources out there, but my impression is that they are written for experienced people, not for beginners in this complex field. In this series, we tried to do something close to real-world kernel exploitation, but much simpler, so people can use it as a first contact with Linux kernel exploitation. I hope people like it, and we will be available to help anyone who wants to reproduce all the content.

内核提权议题 @kernullist 原文 ↗

签名证书滥用持续发生,登录 VirusTotal 可见网络侧检测

Signed "FILIP KOZJAK" and added to the Cert Graveyard. Regular reminder to be logged into VirusTotal even if it is a free account: If you're logged in, it shows you Proofpoint's Suricata detection for the file reaching out to Telegram. https://t.co/aW35kqvZnJ

供应链签名事件 @SquiblydooBlog 原文 ↗

单口硬件整合多类无线攻击,便携红队工具演进方向

RT @therealshodan: Two years ago I released the USB Army Knife, this combined WiFi, BTLE &amp; USB attacks into a single dongle with a unified language. Two years of long nights later I’m teasing the UAK Pro More WiFi &amp; BTLE attacks, mic, IR TX, mesh networking and more (see 🧵) Coming soon!!! https://t.co/F0hXBb6Y3v

硬件工具无线 @cyb3rops 原文 ↗

🪲 Learn Malware Analysis 👉 https://t.co/RfWfShJJX3 Unpacking VMProtect Unpacking Packed Files Series Setting Up CAPEV2 Sandbox Unpacking PECompact Practical Malware Analysis Exercises Unpacking FlawedAmmyy Unpacking ASPack Malware Analysis VM Setup Top 5 IDA Pro Plugins for Malware Gootkit Decryption with Python Removing Obfuscation in IDA Pro Top 5 Malware Analysis Websites LimeCrypter Analysis: Crypter Internals Reverse Engineering Go Binaries StormKitty C# Stealer Analysis Rebuilding IAT for Malware Analysis Unpacking Ramnit Dropper Best SysInternals Tools for Malware Writing YARA Rules ChatGPT for Malware Analysis Dumping Malware at OEP CyberChef for Malware Analysis Downloading New Malware Samples CyberChef: DCRat Loader Analysis Dumpulator for Malware Analysis Miasm Python RE Framework Intro API Monitor for Malware Analysis Process Monitor for Malware Analysis BlackNET C2 with FakeNet-NG Unpacking YouHacker & Python Malware Beginner Malware Traffic Analysis VKeylogger Analysis: Dumping & Fixing Imports Beginner Malware CTF (CyberDefenders RE101) Miasm: Symbolic Execution Finding Malware C2 Panels Malware Persistence Techniques Redline Infostealer: C2 Analysis Mallox Ransomware Analysis Windows Defender Bypass (Token Manipulation) Binary Refinery Tutorial WhiteSnake Stealer Analysis CAPA for Malware Analysis Comparing Binaries with BinDiff APT37 RokRAT Analysis (North Korea) Learning Malware Analysis (Beginners) Paradies Clipper: Crypto Jacker Analysis Reverse Engineering Skid Malware CrashedTech Loader Analysis Fileless Malware on Linux via Scripting File Type Identification Tools KLBanker String Decryption with Python Miasm: Basic Syntax Detecting Manual Mapping in Memory BlackGuard Infostealer Analysis Malware Analysis Learning Resources Miasm: Jitter PE-sieve for Malware Analysis Detecting Process Hollowing Finding Shellcode in Malware Memory LockBit Ransomware KillChain LockBit: Extracting Binary from XLL PolyGlot Files & IcedID Evasion Malicious OneNote Documents (AsyncRAT) Popular UAC Bypass Techniques VFlooder: VirusTotal Flooder Analysis Popular Windows Malware Analysis Tools ProCDOT for Malware Analysis pySMT Deobfuscation via SMT Solver Back-Doored RedLine Targeting Skids CapLoader for PCAP Analysis RedLine Targeting YouTubers Analyzing Office File Malware (Oletools) Top 7 Malware Detection Techniques Regshot for Malware Analysis Detecting DLL Side-Loading Detecting C2 Servers Scanning Memory for Malware Threat Intelligence in Malware Analysis DoNex Ransomware Analysis (Malcore) IcedID / BokBot Banking Trojan Analysis Analyzing Malware Stager Techniques

malware议题 @GuidedHacking 原文 ↗

RT @scaling01: We haven't seen a new chinese frontier model in a while but given what we have seen from GLM-5.3 and Kimi-K3 we should expect some > 160 ECI, but < 165 ECI models before the end of the year and sometime between Feb and Nov 2027 we should see chinese models as capable as GPT-6 Astra and Opus 5.5, likely around May or June 2027 this is when I expect things to go nuts. things like: - rogue agents - cyber incidents - message boards but I'm reasonably sure that chinese models will interact with the outside world in unintended ways before their release (which is how we will first learn about them). So we might already see some funky things in Q1 2027

llm预测 @0xocdsec 原文 ↗

I need you guys to help me work through this conflict that I'm having about Mac Studios. I would love to wait and order the 512 GB variant of the M5 Ultra Studio and I think what they've built is pretty amazing. The problem, I think, that these larger devices have is that if you're getting 512 GB, you are presumably going to run some of the largest available models. The problem with that is the largest models generally have far more active parameters, which also means it's going to run much slower. Almost unusable imo. I'm not trying to use 30tps. So then the question becomes, is it really worth it to run a model with 40B active params? And if not, then why would you need more capacity? Is the 256GB not the sweet spot with a model like Qwen 3.8 Next? Only 6B active parameters is going to run far better on Apple silicon. I'm not convinced the trade off is worth it. Especially when you factor in concurrency. Thoughts?

本地部署硬件 @LLMJunky 原文 ↗

移动端本地数据库无法自证完整性,基于客户端记录取证的结论需重新审视。

RT @TalBeerySec: Burning a possible @BlackHatEvents submission just to make a point: I modified WhatsApp to permanently rewrite its own chat history by editing the local message DB. Kill the app, reopen it — the altered history is still there. For good. Even the DB cannot prove authenticity.

移动端取证事件 @cyb3rops 原文 ↗

RT @rawsalerts: 🚨 #BREAKING: OpenAI warns employees that spies could be trying to seduce them; women may not be with them for their personalities. 📍Silicon Valley, CA “If she’s a 6 or above in San Francisco she’s a spy” warned one commentator.

内部威胁ai_agent @0xocdsec 原文 ↗

0 检出的远程控制植入体,同架构固件设备存在持久化风险需排查

RT @VirLabs_AI: New VirLabs research: Tessera, an undocumented backdoor targeting MikroTik RouterOS on TILE-Gx. The samples had 0 detections on VirusTotal. Underneath a small launcher and wrapper sits a fairly sophisticated remote-control implant. https://t.co/Ak3KMhc5yt

后门固件事件恶意软件 @0xocdsec 原文 ↗

赏金收益全部来自闭源模型,对选型与漏洞挖掘自动化有直接参考价值

takeaways are: - Astra is one beast of a model (when no refusals). It is almost scary. - Open wieghts models have a lot to catch up on - Use more than one model to get better coverage (helped me score 6 digits in bounties $$$) - None of my bounties came from openweight models

llm漏洞挖掘报告 @0xAsm0d3us 原文 ↗

聚焦攻击者滥用合法工具(LOL*)的检测工程思路与共享知识库

RT @alehurtadoxo: I had two of the founding fathers of detection engineering on Detection Dispatch to talk about what they've been farming with the community. 🌱 LOLDrivers. LOLRMM. LOLSkills. Catalogs that track legitimate tools and components that attackers abuse, giving us a shared place to build knowledge, hunt, and detection engineer the ways of prevention.....but the conversation went somewhere I didn't expect. I've spent so much of my career thinking about detection that this episode genuinely had me rewiring my brain around app control/prevention....Prevention isn't a replacement for detection. We need both, and I think we need to talk a lot more about how they work together. That's the episode, you don't want to miss. Thank you @M_haggis & @_josehelps for the laughs 🦉 🦉 🦉 🦉 Full episode dropping this weekend, subscribe to my Youtube channel to be notified 🔔

检测工具议题 @ipurple 原文 ↗

展示 Agent 通过 MCP 驱动 IDA 完成反虚拟化的可行性,逆向自动化新范式

RT @in19h: Devirtualized Easy Anti-Cheat (EAC) for Linux via Gemini 4 Argon and GPT 6.1 Sol using Hex-Rays IDA Assist's MCP: https://t.co/ijFjisPpQe https://t.co/ProwhKO9DZ

逆向ai_agentmcp反作弊 @0xocdsec 原文 ↗

覆盖内核利用、EDR 内部与绕过成体系的免费资料,适合系统补课

RT @cr3ghost: If you're into red teaming, Windows internals or reverse engineering, bookmark this. Adam Chester (@_xpn_ ) has built one of the best FREE offensive security research archives online. The rabbit hole covers: - Windows internals - Low-level reverse engineering - Kernel exploitation - Exploit development + ROP - EDR / AV internals - EDR evasion research - C2 development + tradecraft - Cobalt Strike internals - Active Directory + Kerberos - LAPS internals - Okta + Entra ID + ADFS - Identity provider post-exploitation - SCCM - ETW + .NET internals - Mimikatz internals - macOS security + Mach-O loaders - AWS + Azure - Malware analysis - Red team infrastructure - AI-assisted offensive security - LLM-generated Mythic agents Some absolute classics in here: Kerberos attacks explained under the hood, Mimikatz WDigest + SSP internals, Object Overloading, LAPS 2.0 reversing, identity providers for red teamers, custom Mach-O loaders, Dev Tunnels as C2, and now LLM-driven EDR reverse engineering. This is the kind of archive where one article turns into a 3-hour rabbit hole. Start digging: https://t.co/eJG4TWhxaM Huge credit to @_xpn_ for publishing years of deep technical research publicly. #WindowsInternals #RedTeam #ReverseEngineering

内核免杀逆向工具 @cr3ghost 原文 ↗

RT @nickcammarata: interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works https://t.co/zHij5zwKGT

llm议题 @FuzzySec 原文 ↗

滥用 AI 品牌 + ClickFix 手法,静态免杀明显,需行为与沙箱检测。

⚠️ 𝟱 𝗺𝗼𝗻𝘁𝗵𝘀 𝗼𝗹𝗱. 𝟬/𝟵𝟮 𝗱𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻𝘀. 𝗦𝘁𝗶𝗹𝗹 𝗺𝗮𝗹𝗶𝗰𝗶𝗼𝘂𝘀☠️ ​ https://t.co/LElYRAAzfB researchers uncovered a fake Claude installation campaign using 𝗖𝗹𝗶𝗰𝗸𝗙𝗶𝘅-𝘀𝘁𝘆𝗹𝗲 𝗣𝗼𝘄𝗲𝗿𝗦𝗵𝗲𝗹𝗹 𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻, despite the domain shown in the screenshot receiving 𝗻𝗼 𝗳𝗹𝗮𝗴𝘀 𝗳𝗿𝗼𝗺 𝟵𝟮 𝗩𝗶𝗿𝘂𝘀𝗧𝗼𝘁𝗮𝗹 𝘀𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝘃𝗲𝗻𝗱𝗼𝗿𝘀. ​ A timely reminder: "𝗻𝗼𝘁 𝗱𝗲𝘁𝗲𝗰𝘁𝗲𝗱” 𝗱𝗼𝗲𝘀 𝗻𝗼𝘁 𝗺𝗲𝗮𝗻 “𝘀𝗮𝗳𝗲.” ​ Behavioral analysis and sandbox detonation remain critical when attackers abuse trusted AI brands and legitimate-looking installation workflows. https://t.co/sevAnNcLBH ​ #ClickFix #FakeAI

event恶意软件绕过 @0x534c 原文 ↗

自托管开源方案,可统一管理自有 Linux 设备的配置与合规。

Linux deserves the same care every other platform gets. lindm v0.1.0: a self-hosted, open-source MDM for Linux fleets. Your server, your devices. https://t.co/oZbtoMnmiT https://t.co/Hz0G6kCe3t

工具云 @rad9800 原文 ↗

首次技术记录广告位投递间谍软件,移动端无交互感染路径值得跟进。

RT @billmarczak: Our latest @citizenlab report highlights a 2024-2025 AdFection attack on news network Al Jazeera that infected 45+ devices with spyware from Integrity Labs (née Candiru). This is the first technical documentation of infection-on-impression via mobile ads: https://t.co/34xGE6tsEt

移动端间谍软件报告0click @0x64616e 原文 ↗

老牌社媒信息收集工具维护更新,可用于账号画像与开源情报收集。

Osintgram — Updated! — Osintgram is a OSINT tool on Instagram. It offers an interactive shell to perform analysis on Instagram account of any users by its nickname https://t.co/iHhJjpaTz8 https://t.co/SgNwMP5c23

工具osint @KitPloit 原文 ↗

补充同一广告投递间谍软件活动的目标面,涉及高价值机构网络。

RT @RonDeibert: We show that Discover-Tech ads were shown on devices in national-security-sensitive networks around the world. Numerous ministries of defense, armed forces, cyber agencies, IGOs, multilateral security orgs, including NATO, the UN, and critical infrastructure https://t.co/tketK6AG4z

移动端广告间谍软件 @0xocdsec 原文 ↗

RT @jsrailton: 🚨NEW INVESTIGATION: zero-click mercenary spyware hacking via targeted ads. A normal-looking ad loads. If you're a target, you're infected. No click needed. To protect yourself, block ads now. First time it's been confirmed in the wild. We call it #AdFection. Candiru spyware infections at @AJEnglish. 45+ devices infected. Same ad pipeline operated by Israeli company Discover-Tech reached NATO, European Parliament, dozens of gov networks & critical infrastructure incl US nuclear giant Westinghouse. Confiant, which worked with us, found signs ads were displayed at US @DeptofWar, IAEA & ICC. Investigation ongoing 1/ READ our report by @billmarczak @WolfieChristl, me Astrid Perry, Siena Anstis & Ron Deibert : https://t.co/kU2kFMTqjY Confiant's blog: https://t.co/pX8CWAKxTi

间谍软件恶意广告 @NetAskari 原文 ↗

RT @jsrailton: 4/ Billions of their own bidding records show Discover-Tech's malicious ads showed on devices inside networks @NATO to the @UN, defense ministries across EU, parliaments, offices of heads of state. Election commissions. Networks in critical infrastructure: grids, rail, ports, nuclear. Media orgs in 8 countries and counting. Chip fabs... Key: an ad impression ≠ infection, but the pipeline was all over government networks. Let's dig in.

恶意广告间谍软件 @0xocdsec 原文 ↗

Deepseek just granted everyone their balance back from the last 7 days due to a 3 hour outage. This maybe due to a new model they are gonna release possibly Deepseek V4.1 Pro. W response to the outage by giving everyone's API spending back though 😎 https://t.co/kY5Z7vZbdd

llm事件 @weezerOSINT 原文 ↗

新加载器用并行进程逐级传递原语,是可复现的恶意代码加载手法。

Built mainly on a well known Darksword/Coruna workflow, named here 'Godkiller', it also provided two new loader: 'Liangri_26' and 'Nextgen'. Especially the later running a set of parallel processes to get 'primitives' to hand down the chain. 2/3 https://t.co/EqzDd1r3CB

移动端loader报告 @NetAskari 原文 ↗

对 Coruna 衍生框架是否已跨过限制给出了第一手取证判断。

On the Chinese market Darksword/Coruna inspired C2 frameworks are still proliferating and people are very motivated to try to expand them past its limitations, but the big 'win' hasn't been achieved yet. Could 'Divine Seagull' be the breakthrough ? Thanks to @Huntio 'attack capture' we got access to its source code. Come over to our blog to read about our little investigation, for free: https://t.co/W3sPGg4An3

移动端c2报告 @NetAskari 原文 ↗

给出开源模型在漏洞挖掘任务上的横向数据,可作本地选型参考。

@_jensec These are best results for each model+effort. Surprisingly glm 5.3 low found more vulns after two runs. Yes, two runs were done for each model + effort setting and cumulative findings were taken as final results for both runs.

llmai_agent报告 @0xAsm0d3us 原文 ↗

完整利用链拆解,涵盖从漏洞挖掘到渲染器逃逸的方法论。

RT @7h3h4ckv157: Zero to Calc in Six Days: Inside XBOW’s Chrome Full Chain Author: B. DOLAN-GAVITT Source: https://t.co/DwSFwTCCmY https://t.co/uDNZSoczNY

浏览器逃逸报告议题 @J3rge 原文 ↗

iOS 非活跃自动重启这一防护被绕过,移动端威胁模型需重估。

RT @notdan: EXPLOITATION OF SEIZED IPHONES? YUP U DIED. https://t.co/NluuXhAJ2Y Investigators' tools are advancing, and quickly. In recent months, @Cellebrite and GrayKey, two companies specializing in lаmе аѕ fuсk extraction software used by law enforcement, have new capabilities.

移动端ios事件 @0xocdsec 原文 ↗

推理全程本地不上云,是本地 AI 结合屏幕数据的现成样例。

You remember seeing it. You don’t remember where. Screenlogger remembers. 0.2.0 brings a fresh UI, less disk usage, and search powered by local @googlegemma embeddings. An always-on screen recorder for Mac. Describe what you saw. Find it again. All on your Mac. https://t.co/gPFeI0js7F

工具llm隐私 @rad9800 原文 ↗

十几美元硬件即可被动监听 Wi-Fi/Zigbee 频谱,适合无线侧侦察。

RT @Ready4Sushi: Etherbulb beta release for the ESP32-C3 E27 LED bulb and the SP530 RGB LED controller. Your room is full of radio you never see: Wi-Fi, a phone talking to its earbuds, a smart plug on Zigbee, the cell tower down the road, an FPV drone in the park. Etherbulb turns a 15-dollar smart bulb into a lamp that shows you the air. What it does • Uses the ESP32-C3's Wi-Fi radio as a receive-only SDR (the ESP-SDR path found by ESPARGOS): 80 MS/s I/Q, 25 snapshots a second, 256-point FFT on the chip itself • Hears 2.13–2.73 GHz: all Wi-Fi channels, Bluetooth, Zigbee/Thread, analog FPV video, LTE/5G Band 1, 7 and 38 • Classifies signals by shape* (early alpha) • Blue for Bluetooth, green for Wi-Fi, any colour you pick for anything else. Stronger signal, brighter light • Colour matrix in the dashboard: one row per signal, click a colour, set brightness and sensitivity, or drag a range on the live waterfall to make your own rule • 26 presets from ExpressLRS to microwave ovens, plus 412 mobile-network blocks for 39 European countries, uplink and downlink • Web dashboard served by the bulb: live waterfall, spectrum, log with router time, light and dark theme, works on a phone • Wi-Fi and listening at the same time: the bulb stays on your network while it listens • Idle light: a faint glow until its signal shows up • Switch it off and on 3× for 30 min of Wi-Fi, 7× for a rescue network with the update page. After three crashes it boots the previous firmware by itself LED strips too The same firmware runs on the BTF SP530E strip controller (also ESP32-C3): five outputs, five strips, one per signal type, all on at once. One LED per MHz if you like. Hardware •12 W RGBCW E27 bulb, ESP32-C3 inside. Flashes over Wi-Fi, no screwdriver • SP530E: first flash over UART, Open source, GPL-3.0Your Wi-Fi password stays on the bulb. https://t.co/5MMGCcGDl8 #ESP32 #SDR #smarthome #opensource #RF #maker

工具固件无线 @0xocdsec 原文 ↗

CVE 编号清单已出,可据此对照自身补丁节奏。

RT @heegong123: 8 vulnerabilities I reported were patched in this Patch Tuesday: 1. CVE-2026-96243 2. CVE-2026-88070 3. CVE-2026-83586 4. CVE-2026-83877 5. CVE-2026-96911 6. CVE-2026-85791 7. CVE-2026-83905 8. CVE-2026-85909 Hope the bounty review wraps up soon! Thank you! @msftsecresponse

cve微软补丁 @0xocdsec 原文 ↗

ESXi 是自托管虚拟化的常见底座,新产品线的后续细节值得跟。

RT @ericnipro: Announcing VMware HyperArmor, a new solution to help further protect the ESX hypervisor component of VCF. HyperArmor works alongside broader security capabilities provided by VCF, vDefend, and Avi Load Balancer. Interesting. .love the name -eric https://t.co/M9WOY42894 https://t.co/rukYiG2s83

虚拟化云工具 @kernullist 原文 ↗