⚡Megatron LLM Hub
All digests
Twitter

推特安全流

Top8

自建 Tor 中继会被直接压垮,文中给出具体攻击手法与时间线。

Tor’s relay network is under a sustained circuit-creation DDoS. The attack floods volunteer relays with large numbers of client circuits, exhausting CPU, memory, and connection tables. It has been active in waves since at least late October 2026, with earlier related pressure in June through July. The dominant techniques are high-volume circuit builds and BEGIN_DIR floods. BEGIN_DIR is the cell type used to request directory documents (consensus, descriptors) over an already-built circuit. Attackers open many circuits and issue these requests at rates that overwhelm the relay’s processing queue. Some variants produce extreme asymmetric traffic (high ingress or high egress), others drive tens or hundreds of thousands of connection attempts, and still others cause the relay to thrash while reporting relatively modest bandwidth. Operators have observed the method change within minutes of a mitigation or upgrade. HSDir relays (those that store onion-service descriptors) have been particularly affected. Many lost the HSDir flag under load; some operators reported the attack on a given relay stopped the moment the flag was lost. Guard relays have also been hit. The result is degraded performance, temporary unreachability of some onion services, and reduced network capacity, though user anonymity itself is not reported as broken. This is a resource-exhaustion attack on the relays, not a deanonymization attack. On 13 November the Tor Project released 0.4.9.17, an emergency update. The key change (TROVE-2026-031) alters IPv6 DoS accounting: instead of tracking every individual IPv6 address, relays now group clients by /64 prefix. Additional connection-isolation and data-validation fixes were included. The project stated that defending against the flood is not straightforward and that further releases should be expected. Relay operators are advised to upgrade immediately. The attacker has already demonstrated adaptation. Multiple operators reported that after the 0.4.9.17 upgrade the traffic pattern shifted to a different technique within roughly ten minutes. Earlier defenses aimed at BEGIN_DIR and HSDir rate limits were also partially bypassed. No public attribution or claim of responsibility has been made. The network continues to function because of roughly 9,000 volunteer relays, but the ongoing adaptive pressure is forcing repeated emergency releases and is visibly reducing the number of relays carrying certain flags.

ddostor事件 @0xpwnie source ↗

面向中继运营者的补丁,且升级后攻击者约十分钟即换手法。

RT @IntCyberDigest: ‼️ The Tor anonymity network is under a sustained DDoS attack that has hit most of its volunteer relay operators. Attackers are flooding the relays, the servers that carry Tor traffic, with client circuits. The Tor Project pushed out a security release with a key fix to its DoS defences for IPv6, and says more are coming. One relay operator says the attack switched to a new method about 10 minutes after upgrading.

torddos补丁 @0xpwnie source ↗

部分工具有外联回传、带攻击签名、遗留开放shell,自托管前必须知道。

RT @andpalmier: I read the source code and full git history of 14 open-source AI pentest tools (Strix, Shannon, PentAGI, PentestGPT, HexStrike, CAI, Cairn...) and ran 7 of them in a lab. Some phone home. Some sign their attacks. Some leave a shell open for anyone 🧵 https://t.co/WOQFhhDPKG

ai_agentllm工具供应链 @0xocdsec source ↗

已落地的 Agent 化云凭据窃取链,密钥一暴露就等于整片云环境失守

RT @_josehelps: Microsoft Threat Intelligence watched threat actors use the agentic framework ARTEX to search public GitHub repos for exposed secrets, then abuse those secrets against Entra ID and Azure. In one case an AI agent went from finding a client secret on GitHub to pulling data out of Azure Key Vault in five minutes. https://t.co/VMPhtg2Mb9

ai_agent云凭据泄露 @cyb3rops source ↗

补丁未进入主线,PoC 可直接复现,自托管 Linux 主机影响面大

RT @Nadsec11: New Linux Kernel LPE :) Found while preparing for KernelCTF a few weeks back, wouldn't have been a particularly outstanding submission for KernelCTF, woulda got smoked, currently unpatched, but a patch exists in one of Linus's forks. Also no CVE currently assigned. Demo here: https://t.co/pSVpMV34I3 POC here: https://t.co/zjymtcW3ip Patched here: https://t.co/CXkjlFYMBB

lpe内核poc @0xocdsec source ↗

可本地部署的自主攻击 Agent 完成全链路,直接对应 Agent 滥用与自动化攻击风险。

redamon v7.8.0 — Autonomous AI red team framework that chains reconnaissance, exploitation, and post-exploitation into a single pipeline, then triages findings, implements code fixes, and opens pull requests. https://t.co/lxLel1pZ6r https://t.co/ufuYnAqZmZ

ai_agent工具渗透测试 @KitPloit source ↗
Must-see10

公开的半成品内核 R/W 原语,是移动端提权链的关键一环。

RT @hazuk1_Aki: i published an unfinished kernel R/W exploit for iOS 27.0.1. if you think you can get it working, please take a look ❤️ https://t.co/aMNWEtDpuz

poc内核提权移动端 @0xocdsec source ↗

同网段即可能被攻击,涉及 deauth 与恶意 AP 两种路径,细节待公布

RT @Neodyme: What's up, Breakpoint! 👋 It's badge pickup time, so we thought we'd take a closer look at their security. We achieved zero-click RCE against any device connected to a Wi-Fi network (via deauthentication attacks or simply by joining a vulnerable AP). Details coming soon 👀 For a live PoC, hit us up at the venue!

rce移动端poc @0xocdsec source ↗

内核与文件系统驱动提权,是补丁前本地提权的常规入口

RT @malware_owl: Fun side project with @ultimat3hg and Assistant Dr. Claude a number of months back and these are some new updates from @msftsecresponse: October Patch Tuesday 1) CVE-2026-83835 Windows Common Log File System (CLFS) Information Disclosure 2) CVE-2026-96947 Windows Universal Disk Format File System Driver (UDFS) Elevation of Privilege 3) CVE-2026-70121 Windows Kernel Elevation of Privilege Vulnerability July Patch Tuesday 4) CVE-2026-49177 Windows TCP/IP Information Disclosure Vulnerability Can't make any promises, but hopefully I can find some time to do a writeup on one of the more interesting ones (variant analysis). https://t.co/sYfBZgZBi5 https://t.co/WLUfl5tKJp https://t.co/lP54j39Q5b https://t.co/vlb6xp8R3a

cve提权内核 @0xocdsec source ↗

可直接替换 Playwright,通过主流机器人检测,抓取与自动化测试都能直接上手。

invisible_playwright v0.52.0 — Patched Firefox browser with C++-level fingerprint spoofing and humanized mouse automation. Drop-in Playwright replacement that passes all major bot detection suites for stealth web scraping. https://t.co/HKiYZWsGXP https://t.co/fQaQAkeaGB

工具浏览器绕过 @KitPloit source ↗
Recommended15

模块化规避型植入的工程实现,可直接借鉴到载荷开发。

RT @ipurple: A Modern Windows Implant Development Framework for Offensive Engineering and C2 Research. Suture uses Crystal Palace to build a flat binary that pairs a position-independent loader (PIC) with a modular PICO payload. https://t.co/eMs51Wfuk4

工具绕过c2 @0xTriboulet source ↗

逆向主力工具原生接入 Agent,固件基址与对齐判断可自动完成

Coming soon to Hex-Rays IDA Assist: loading a piece of firmware requires knowing the load base and parameters for RAM & ROM alignment; drawing from the zrax classifier, agents can now ask Assist to analyze files before they're even loaded in IDA to inform their interpretation. https://t.co/rpx8cKTruR

ai_agent逆向工具 @in19h source ↗

把回调、交叉视图与句柄检查做成可导航面板,降低内核取证门槛

I've been slacking off a bit so progress has been pretty slow, but I think I've got a rough version working for now. KNDbgX is a WinDbg extension for visual kernel analysis. It supports callback and minifilter inspection, process cross-view checks, handle access analysis, and kernel snapshot comparisons. Results are kept in tabs, with direct navigation to memory and disassembly. repo: https://t.co/hsaKbPXREY

内核工具windows @kernullist source ↗

从机制层面讲清 LDT 与微架构偏差,可作内核漏洞研究参考

RT @leo_s0mething: Just a 2-line patch in the linux kernel... 👀 New CTF challenge writeup by me and @___prosti In the blog-post you can read up on LDT internals, as well as a small microarch discrepancy! Is this yet another x86 backdoor @xoreaxeaxeax ?? https://t.co/gq06o83Tky

内核议题绕过 @0xocdsec source ↗

可自托管的开源移动 App 安全测试框架,能接入 CI/CD 流水线

Mobile-Security-Framework-MobSF v4.6.6 — Automated mobile application security testing framework for Android, iOS, and Windows. Performs static and dynamic analysis, malware detection, and API security testing with CI/CD integration. https://t.co/hGvaJZ06v8 https://t.co/qp6t6ICklc

工具移动端 @KitPloit source ↗

改两处设置即可阻断 2G 降级窃听与基站定位,成本极低

RT @sekurlsa_pw: Android: To disable 2G, switch 2G protection to on. With Android’s Advanced Protection 2G will be disabled. iOS: Lockdown mode will disable 2G and 3G. If an attacker can downgrade your phone to 2G it can intercept some communication and geolocate your device. https://t.co/qvxKwl2NL7

移动端隐私 @0x64616e source ↗

免安装的自托管逆向环境,功能向 Ghidra 符号树看齐,适合快速取证分析。

RT @oneandonlyhusam: New OpenBin update is live! A batch of quality-of-life upgrades for anyone reversing binaries in the browser. 1/ Function tree. The endless function list is gone. Functions, Imports and Thunks now live in folders, and big groups split automatically by name prefix (FUN_0010…, ieee754_…), like Ghidra's Symbol Tree. 2/ Data XREFs. Open any data symbol and see the functions that reference it. Click one to jump straight to the exact line. 3/ Program Listing. The full assembly listing is now syntax-colored, and function names, data labels and addresses in it are clickable. 4/ Pseudocode. Ghidra's types, parameters, locals, globals and labels each get their own color, so decompiled code is much easier to scan. 5/ Also: the left sidebar is resizable, and collaborators on shared npm/script projects can now open them (they were getting a 404). Try it: https://t.co/OzGP5VPT7A https://t.co/ijW7hNtzQH

工具逆向浏览器 @0xpwnie source ↗
More50

跨平台多版本实测 Lockdown Mode 防护边界,可直接对照配置。

RT @naehrdine: We got two publications accepted at CCS, to be presented next week 🎉 Blue Bubbles, Red Flags: Investigating Privacy Leakage in Apple iMessage https://t.co/KQqNzqy7kV Dissecting Apple's Lockdown Mode: A Cross-Platform, Multi-Version Analysis https://t.co/GVoDRyrqwy

移动端隐私报告 @alkalinesec source ↗

勾勒自动化逆向 Agent 的能力边界与可落地任务类型。

What’s the most unreasonable thing you could ask an AI RE assistant to do? Assume Assist can support anything. The more absurdly complex, the better. Not devirtualization or app decompilation. Think bigger. If I find it worth doing, I might build it.

ai_agent逆向llm @in19h source ↗

API 层路径遍历可直接越权读写,自建同类平台可照此自查。

RT @OpenBinAI: We'd like to give a special thanks to @athm32 who found and privately reported a critical path traversal bug in the platform API. Fixed as of last week and the final updates were pushed to day. THANK YOU for keeping https://t.co/Sbxod6EELI SAFE! Please give him a follow!

路径遍历api事件 @0xpwnie source ↗

RT @michael__ni: I am the main developer fixing security issues in FFmpeg. I have fixed over 2700 google oss fuzz issues. I have fixed most of the BIGSLEEP issues. And i disagree with the comments @ffmpeg (Kieran) has made about google. From all companies, google has been the most helpfull & nice

开源漏洞修复 @FuzzySec source ↗

Idk I feel like breaking and porting GTA6 is just a question of someone with a rooted box and some tokens. Assuming infected, I'd probably install a fresh Windows10 on a separate SSD with an own EFI and disable all internet traffic. Since other disks should be encrypted np.

逆向游戏 @0xocdsec source ↗

RT @NathanMcNulty: My 12 year old fully set up my R630 (28 core, 384GB RAM, 4TB NVMe, 8 TB disk) with Proxmox, vlans, and AD using Luna 6 on Max on my OAI sub, used hardly any quota I gave her tips to ask Luna, like ask about backups, ask about vlans, etc., but the future is freaking awesome 😎 https://t.co/W39tQNKdY8

llmai_agent运维 @bohops source ↗

Do you want to become from [ OPSEC level: lack opsec skill ] to OPSEC level: GOD MODE. Then you should read: EXTREME PRIVACY: WHAT IT TAKES TO DISAPPEAR. i am reading this book for a while now. its an great one. https://t.co/zQGnlXJqbK

opsec书籍 @5mukx source ↗

多租户隔离绕过与SaltStack攻击链的可复用手法复盘。

RT @kl_secservices: In a pentest of a major cloud provider, we started as a regular tenant. We chained several bugs to break tenant isolation and reach other customers’ Kafka, Postgres and Redis. Then an SQL injection in SaltStack got us into the cloud’s main database. more: https://t.co/GeOCJdBrLf

云sqli提权事件 @0xocdsec source ↗

硬件级绕过微码信任根的思路,涉及所有现代x86平台的启动信任链。

The microcode patch RAM on modern x86 is loaded before the first instruction the OS can see. Signature verification runs inside the sequencer. A voltage glitch timed to the compare step can force acceptance of an unsigned patch. The CPU continues with the modified behavior and the later cryptographic checks doesn't re-examine the already loaded patch.

固件绕过硬件 @0xpwnie source ↗

后渗透阶段的一站式工具链,可自托管部署验证。

RT @chenbuwan1995: 红队行动拿到初始权限后,怎样继续深入目标环境? Nukerat 是一款面向高级红队行动的后渗透框架,帮助安全团队探索后续行动的更多可能 了解项目:https://t.co/CIWXmWr63W https://t.co/bU6CjTB5oZ

工具提权红队 @0xocdsec source ↗

UAV攻击面有现成可复现靶场,适合动手练习。

RT @DI0256: Drone Hacking: Hacking UAVs with Damn Vulnerable Drone https://t.co/afKrrKWWiT For more on this topic, including the practice, join the upcoming Hackers-Arise online training session on NOVEMBER 17–19! https://t.co/Upq31jMKFr

工具固件poc @0xocdsec source ↗

未保存 RSP 就对齐栈会让 ret 跳到零地址,是 shellcode 崩溃的常见根因

RT @RubberDuckCz: Win x64 Shellcode Part 4 is out. Calling convention, shadow space, stack alignment. And why 'and rsp, 0xFFFFFFFFFFFFFFF0' without saving RSP first causes ret to load zero into RIP. https://t.co/GLjSSgBG7c #shellcode #windows #x64 #asm #redteam https://t.co/part8tEMMC

shellcodewindowsredteam @ipurple source ↗

Following along with the (excellent!) custom hypervisor series for 'fun' and learning (porting the original series to Rust ofc ^^, and trying to make it somewhat idiomatic) has been enjoyable so far, fairly slow going; but making progress! A few Rust challenges along the way, most notably language wise porting union bitfields. The first one I came across in anger I masked directly out of the bitfield into a struct. No problem. Then I came to part 4 of the series for allocating the EPT, and felt my will to live get drained lol (lots of bitfield unions :E). Came across a bitfield crate which makes make life a LOT easier..! I'm going to have to read through part 4 of the series a few more times. Gonna slow cook a Sunday stew and let everything up until now 'stew' in my head...

虚拟化rust @0xfluxsec source ↗

RT @Defence_Index: 🇹🇷 BREAKING: Turkey’s new drone killer just took out 51 drones without firing a single missile or bullet. ASELSAN’s EJDERHA uses powerful electromagnetic waves to disable drone electronics. The system has entered serial production, adding a new dimension to Turkey’s counter drone warfare capabilities. https://t.co/VrJIcTufxX

硬件事件 @0xocdsec source ↗

RT @ChrisO_wiki: 6/ While hiSky is technically inferior to Starlink, with speeds of 7 to 2000 kbps compared to Starlink's tens or hundreds of megabits per second, this is said to be sufficient for use as a "digital leash" – for sending short telemetry packets and receiving commands. https://t.co/0Lk2d6Hsqj

无线电事件 @0xocdsec source ↗

RT @TheQuantClips: Former Citadel head of equity quant research Dan Morillo on why AI can't generate alpha yet: "It is trained on millions of documents, thousands of millions of documents." "In finance, you would think of this as, it gives you the consensus." "If you put a prompt in ChatGPT, what you get back is essentially some average of what a typical person might say." "When you think about alpha generation, you don't make money off of the consensus. You make money off of differences." "We just don't think it's at that stage at this point."

llm @0xTriboulet source ↗

监控盲区与假冗余的典型反例,自建服务的架构自查可直接对照。

RT @GergelyOrosz: Revolut shared the postmortem of their global cards outage of 5 hours. Some interesting (alarming?) notes: 1. They had no monitoring on database nodes going down (!!) 2. Across 3 regions, they ran 5 nodes, with #1. So a single node going down meant no zonal resillience (cont'd) https://t.co/XG8ocvmPKC

可用性云故障复盘 @cyb3rops source ↗

给漏洞优先级判断提供口径:按可利用性而非数量排序。

RT @justdionysus: @halvarflake @HackingLZ @TalBeerySec @natashenka I think vulns are dense in most useful software. I also think reliably exploitable vulns in critical surface are sparse. Fixing those are useful (as part of a balanced seceng response). GPZ is reporting those. That’s also what Calif tries to report.

议题报告 @alkalinesec source ↗