⚡Megatron LLM Hub
All digests
Twitter

推特安全流

Top8

自主 Agent 缺少边界约束属共性风险,直接影响自托管部署。

In her latest piece for @axios, @samsabin923 reports on security agent scope adherence and efforts to keep them under control. Timely topic following our release of #ScopeBench last week at @OffensiveAIcon, which is referenced in the article alongside @SpecterOps' research. Check it out! https://t.co/TthUpOH2s5

ai_agentllm报告 @dreadnode source ↗

无需认证即拿服务器最高权限,已有在野利用并有完整技术分析

Check out our latest deep-dive for #AhsayCBS, an enterprise backup solution. Reported as exploited in-the-wild by @HuntressLabs, CVE-2026-105133 and CVE-2026-105134 allow an unauthenticated attacker to execute commands as SYSTEM or root on the server. https://t.co/WsoHNh32BN

rcecve事件备份 @Horizon3Attack source ↗

完整全链利用(含沙箱逃逸)细节公开,复现路径与技术门槛明确。

Ask @moyix anything about our Chrome full-chain exploit. 👀 He published the technical breakdown of how he did it (https://t.co/64pCOSumuu)...but if you have questions or want to go deeper, drop your questions in this thread! He’ll be answering them LIVE on Wednesday, 10/14, from 12–1 PM ET. 💚

浏览器rcepoc报告 @Xbow source ↗
Must-see8

EVM/Solana/TRON/BTC 同时失血,可复盘多链攻击与资金追踪路径。

RT @vxdb: Full picture of the Gamdom exploit ~2.2M stolen on EVM (ETH, BSC, POL, ARB, OP) 965k stolen on Solana 375k taken from TRON and 668k on BTC as well https://t.co/EKEOrRGIsK

事件加密货币攻击手法 @0xocdsec source ↗

厂商修补未覆盖真实攻击链,最新固件仍可被远程打穿。

RT @satoki00: Pwn2Own本番で失敗したプリンター。Exploitをしっかり調整したら最新版でも刺さりました。ガジェットサーチが適当すぎてバグってたらしい。RCEを二件報告済み😑 #Pwn2Own #P2OIreland https://t.co/uI7Xn8xmuM

rce固件poc @0xocdsec source ↗

LLM 驱动的漏洞挖掘已进入闭源商业软件,披露与修复节奏会随之变化。

RT @ZixuanLi_: GLM just helped uncover 23 Windows vulnerabilities, all acknowledged by Microsoft MSRC. Beyond open-source projects, commercial software security is also being improved through GLM-powered independent vulnerability discovery and responsible disclosure. https://t.co/TsaCL8UBLR

llmcvewindows @0xocdsec source ↗

VBS 是 Windows 关键隔离边界,能在此实现 VTL1 代码复用的公开研究极少

RT @Doyensec: 🔎 New Windows vulnerability discovered by Adrian Denkiewicz (@a_denkiewicz) of #doyensec - a high-severity flaw in one of Windows’ Virtualization-Based Security (VBS) enclaves that enables VTL1 code reuse. A first look at our ongoing research into this class of vulnerabilities. 🔗 https://t.co/l1cmjxc6o6 #appsec #security

windows提权绕过 @0x64616e source ↗

三类设备市面可购、成本低,物理植入后可长期音频监控。

For those new to the audio surveillance game let me intro you to some that I have found or used: - Wall mic abused by your neighbors - Malicious cables with mic & micro SD recording similar to @_MG_'s work - Powerstrip/cables with mic & optional wifi/4g (thanks iSOON leak) https://t.co/JMad4LUihg

硬件窃听报告 @Laughing_Mantis source ↗
Recommended15

大规模真实 Beacon 语料可直接用于检测规则训练,此前无公开等价数据。

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

数据集工具c2 @domchell source ↗

不依赖 SMB 与远程注册表即可抓取凭据,横向隐蔽性提升。

RT @_zblurx: Little upgrade on NetExec WMI protocol 🚀 Now NetExec supports SAM, LSA and NTDS dump using only WMI. Thanks a lot to @al3x_n3ff for all the help on this PR (and all the work on NetExec overall, you rock bro 🫶) https://t.co/z46sT9nfqV

工具凭据windows @pentest_swissky source ↗

避免在假 PoC 上耗费时间,也提示投毒仓库借热度传播的风险。

RT @GenericCoding: Everything which claims to be for 26.x in this repo is fake. Even where an exploit could properly be used on the claimed supported version, they are not properly triggered. For instance the userland PAC bypass could work on 26.1, but in this repo the implementation is wrong.

iospoc伪造 @0xocdsec source ↗

BYOVD 已成产业化的 EDR 致盲手段,量化比例可用于校验防护假设。

Nice read by Sentinel and some interesting numbers like: "Ransomware actors successfully disabled security tools in about 48% of intrusions in 2024." This is a very big ratio. Also, LOLDrivers had its moment. 😀 @M_haggis , @_josehelps @nas_bench https://t.co/Q0zzm1W7mK

报告edr驱动 @ipurple source ↗

同一签名证书被重复滥用,附带的哈希与 C2 可落地为检测与封禁规则。

Low detection Stealer, internally named "Zalmoxis". Uses a signer that we've seen from two other CA: "YOUR CHANCE j.d.o.o." ad0d08d54c9bbe2309bb7a8ae92def8a04297d0238f2799764d93929b57cbd5f https://t.co/a7AdqogjP1 cdn[.]simplynetworking[.]eu Zalmoxis c2: 45[.]152[.]149[.]66:6969 Panel was on port 3000 Previously seen signed as "Komers Trading Company LLC" 1534e8091708e13a24698de848faccac0ccbf82cc625f990bed0c7ec5388c345 h/t to @malwrhunterteam We've seen this certificates for this signer issued by SSL, Sectigo, and GlobalSign

恶意软件签名事件 @SquiblydooBlog source ↗

双签名加载链细节与完整 IOC 可直接转写为检测规则。

h/t @malwrhunterteam Signed EtherRAT dropping signed StealC: EtherRAT: 1b61b95a61996b44322e9841af2fa2b63fcdd18a569f16fa8acd30eb8d8ac854 Signer: "KARA-TASH I KO, OSOO" StealC: 37d7a8c504227f9c1986619fd8148f1737cf94e56ac91687f8ba33be7e945f5e Signer: "AirTiki ApS" C2: presanis[.]com Stealer Download: csg-soft[.]com / 45.135.180[.]185 StealC C2: glamour-studio[.]com / 194.11.226[.]153

恶意软件事件ioc @SquiblydooBlog source ↗

复用微软签名 DLL 承担加密可减少体积与内存特征,红队可直接借鉴。

It has been a quieter month than usual for iPurple blogging, as I have been working on content for a number of upcoming projects and activities. A new article will be published early next week. In the meantime, have a look at my latest post on leveraging WerEnc.dll to encrypt your implant. Why use it? ✅ Smaller stagers by reusing encryption code already built into Windows ✅ AES-256 encryption support ✅ Trusted, signed Microsoft DLL Living-off-the-land encryption has its advantages. https://t.co/JUOyyYHbRb

工具绕过windows @ipurple source ↗

不偷密码而拿 M365 令牌,绕开常规口令防护,诱饵与域名细节齐全

🎣 https://t.co/LElYRAAzfB researchers identified 𝗦𝗸𝘆𝘄𝟰𝗹𝗸𝗲𝗿, a 𝗱𝗲𝘃𝗶𝗰𝗲 𝗰𝗼𝗱𝗲 𝗽𝗵𝗶𝘀𝗵𝗶𝗻𝗴 campaign primarily 𝘁𝗮𝗿𝗴𝗲𝘁𝗶𝗻𝗴 𝗨.𝗦. 𝗼𝗿𝗴𝗮𝗻𝗶𝘇𝗮𝘁𝗶𝗼𝗻𝘀 across the manufacturing, technology, and healthcare sectors. The campaign uses password-protected contract or service-proposal decoys and sites[.]google[.]com lures to redirect victims to Microsoft authentication, seeking M365 token access without directly harvesting passwords. ​ #Skyw4lker #DeviceCodePhishing

钓鱼云事件 @0x534c source ↗

免费成体系的红蓝队资料,覆盖 AD、Entra、Okta 与云多条攻击路径。

If you're into adversary simulation, identity security or red/blue tradecraft, bookmark this. The team at @SpecterOps has built one of the best FREE security research archives online. The rabbit hole covers: - Active Directory + Kerberos - Entra ID + Azure - Okta + identity providers - BloodHound + attack paths - SCCM / Configuration Manager - Red team tradecraft - Detection engineering - Lateral movement - Credential access - OAuth + token security - Browser persistence - AWS + cloud attack paths - Kubernetes security - Software supply chain attacks - Adversary simulation - Purple teaming - Threat hunting - AI-assisted offensive security There is a ridiculous amount of deep technical material here, including research developed from real offensive and defensive operations. They also now have FREE on-demand Tradecraft Academy courses covering BloodHound, Kubernetes for Red Teamers and software supply chain security. Blog: https://t.co/HRO177M5lY Free Tradecraft Academy: https://t.co/5DIk2Mmtwi Huge credit to the @SpecterOps team for publishing so much of this openly. #RedTeam #ActiveDirectory #Infosec

ad云工具报告 @cr3ghost source ↗

免费提供 Elastic/Splunk 合成遥测与已验证调查问题,可练习狩猎。

RT @ThruntingLabs: ✨ 𝗧𝗵𝗿𝗲𝗮𝘁 𝗛𝘂𝗻𝘁𝗶𝗻𝗴 𝗟𝗮𝗯𝘀 𝗩𝟮 𝗶𝘀 𝗹𝗶𝘃𝗲! 𝗡𝗲𝘄: ✅ A redesigned UI/UX across the platform ✅ Tailored Hunts with synthetic telemetry in Elastic/Splunk, 10 validated investigation questions, and full reports. ✅ A query builder in Search Console ✅ Native Azure Logs portal access ✅ New improved Hunt Leads ✅ Team technical tabletops! Details in our blog post: https://t.co/FYW6bLiohg

工具检测报告 @ipurple source ↗
More40

AI 实验室内部窃密现实案例,指向内部威胁与供应链风险。

RT @unusual_whales: BREAKING: The OpenAI employee who was arrested and charged with allegedly using a hidden recording device in one of the ChatGPT maker’s San Francisco offices had a one-way ticket to Saudi Arabia booked for October 12, per Bloomberg.

事件内部威胁ai_agent @0xocdsec source ↗

第三方内核模块进入发行版,扩大内核攻击面并带来权限争议。

BIG DEAL ALERT. Fortnite is coming to Linux, which means a supported kernel level anti-cheat is also coming (to some distros). I know those types of privs are not popular with many people, but this will pave the way for other major titles to finally come to Linux -- and maybe, just MAYBE move the needle when it comes to Windows dominance. Shout out @EpicGames

内核linux反作弊 @LLMJunky source ↗

AI 辅助审计已在被广泛依赖的基础库上产出真实 CVE,修复节奏值得跟踪。

RT @IntCyberDigest: ‼️ European AI security firm Aisle found 24 curl vulnerabilities after Anthropic's Mythos and OpenAI's Codex could no longer find any. curl lead Daniel Stenberg says Aisle has found 35 curl CVEs in about a year, more than any group or person. https://t.co/KWbSAfOwrA

llmcve供应链 @0xocdsec source ↗

若属实影响面覆盖 xterm、Kitty 等通用终端,但目前无细节与 PoC 可验证。

RT @VixWizzer: I discovered a critical privilege escalation vulnerability affecting nearly every terminal emulator (xterm, Kitty, etc.) The payload was just two letters🤯 Instant root on most systems. I’ve already responsibly disclosed it to the maintainers. Waiting for the CVEs.

lpe提权 @0xocdsec source ↗

奖励结构调整会改变浏览器漏洞的提交流向与公开时间线。

RT @zerodayalpha: Chromium VRP just raised reward amounts – after lowering it to near 0 earlier this year. 4x memory safety baseline, increased rewards for logic sandbox, and the fuzzer bonus is back (!) Still no incentives to report the bugs that could pay 10x at comps... https://t.co/Bn1v6XkTPA

浏览器漏洞赏金 @0xocdsec source ↗

RT @batwood011: Selling your company's data for (less than a few years' revenue of) cash is the most retarded, short-termist thing you could do, and everybody seems to be doing it. The more data you sell to an AI lab, the more likely your entire company will be one-shotted by the next model. The labs must think they are getting away with murder, because they essentially are. It's especially goofy given @thinkymachines, @PrimeIntellect and others help you train your own models to keep all the benefits of automation in your business to yourself.

llm供应链观点 @0xTriboulet source ↗

展示个人 Agent 拿到屏幕与语音输入的形态,权限面即数据暴露面

RT @Trevs_Dev: I’ve been working on this over the weekend On the Apple Vision Pro, I made a Ghost and I set it up to work with Codex. It's context aware since I gave it the ability to take screenshots of my screen and voice is enabled, so we have conversations. Here I’m asking it to make a solar system app and diagram (you can’t hear me though) #buildinpublic

ai_agentllm工具 @Laughing_Mantis source ↗

呈现植入式监听设备的真实能力与来源,非理论推演。

I am not the only one who has used these as well, as i mention above the iSOON leak of 2024 gave us a glimpse of threat actors who act like a paid for hire APT with a vast range of capabilities Their Wifi Sesorless Implantable devices with mics allowed them to remotely monitor https://t.co/MbG2G20lqp

硬件窃听事件 @Laughing_Mantis source ↗