⚡Megatron LLM Hub
All digests
Twitter

推特安全流

Top7

补丁已发布且攻击演示公开,未升级的 iOS/macOS 设备直接暴露。

RT @calif_io: Welcome to Hotel Calif. We're in lockdown mode. You can check in, but you can never leave. Broccoli is served around the clock. The Brotli bug apparently affected iOS and macOS too. Apple released version 27.2.1 yesterday, fixing this bug among other security issues. This is an important security update. If you haven't updated yet, now's a good time. The demos show what we were able to do with it.

cve移动端浏览器 @patrickwardle source ↗

直接可上手的 AI 代理执行框架,便于评估其监督模式的风险边界。

RT @SpecterOps: Meet Sage: open-source AI for assumed-breach operations, built into Mythic. State your objective; Sage prepares commands for your approval in supervised mode. @Ne0nd0g breaks it down in his new blog ⤵️ https://t.co/haYwfjObTc

ai_agent工具llm @0xTriboulet source ↗

公开 PoC 复现该规避手法,可据此改进 .NET 样本检测规则

CerealLoader is the first malware I've seen to use custom mixed-mode assemblies that actually split logic between managed and unmanaged code using C++/CLI. By "hiding" execution of some of the malware code behind native stubs (.nep section), this technique makes analysis more challenging. I've written a POC to replicate how it functions: https://t.co/ks67ecPp74

恶意软件绕过poc @TheEnergyStory source ↗

攻击链可复现、上游是否合并未定,AD 环境需按 WPAD+RBCD 检查暴露面。

RT @nullenc0de: Not sure if this will get merged, but I was able to go from unauthenticated access to Domain Admin using a patched version of Impacket that I developed. I believe CVE-2026-70229 should be incorporated into Impacket. Attack chain: WPAD > RBCD > DCSync https://t.co/xst4EVmo7k

提权cvepocwindows @0xocdsec source ↗
Must-see19

“私有应用”的默认假设被证伪,链接即凭据的设计会外泄进 DNS 日志。

RT @IntCyberDigest: ‼️BREAKING: Grok Build has left 35,000+ of its users' apps open to anyone, including apps from SpaceXSI. The open apps held 3,784 medical exams of workers and a Tesla technician scorecard, plus a Navy contractor's roster and SpaceXSI's own sales plans. The "secret" words in each app's link sit in its web address, which DNS logs record, security firm Claroc found. Cursor, which SpaceXSI bought, says apps are private by default. Claroc's tests show otherwise. SpaceXSI pulled its 24 apps after Ars asked. Most other data was still exposed when Ars published.

ai_agent云事件 @0xocdsec source ↗

面向新 SLUB 缓存机制的内核利用链,具备复现条件的研究素材。

RT @linkersec: Harvest Season for SLUB: From io_uring vulnerability to Novel Sheaf-Based Exploitation Techniques Article by @naup96721 and Yu-Ting Lin about the SLUB exploitation techniques applicable with the new SLUB sheaf/barn caching mechanism. https://t.co/KkaRt8Wubc https://t.co/Q4nMPsXJjP

内核提权报告 @0xocdsec source ↗

Outlook 0-day 已在野利用,恶意邮件文件即可泄露 NTLMv2 哈希。

RT @BleepinComputer: ‼️‼️ Microsoft’s November 2026 Patch Tuesday fixes a massive 1,235 vulnerabilities, including 66 rated Critical and an actively exploited Microsoft Outlook zero-day. The Outlook flaw (CVE-2026-124194) can be exploited using malicious message files that trick Windows into connecting to attacker-controlled network locations, leaking NTLMv2 authentication hashes that could be used to impersonate victims. Of the 66 Critical flaws, 48 are remote code execution vulnerabilities, 15 are privilege escalation flaws, and three are use-after-free bugs in Microsoft Edge. The massive update follows October's record-breaking 1,411 vulnerabilities. ➡️Full details: https://t.co/c2NE6TQtwN

cve0dayntlm @artem_i_baranov source ↗

此前未公开的 ESXi 后渗透工具,为虚拟化环境检测提供新线索。

RT @_whoisnt: Excited to share new research I co-authored with @Lawrence_Sec We uncovered GhostFrame, a previously undocumented VMware ESXi post-exploitation framework used by TAG-218, whose activity overlaps with Settra ransomware. Read more: https://t.co/XuY6nEtD84

esxi恶意软件报告 @0xocdsec source ↗

该 API 误用不会报错也不会改写 key,签发证书即等于泄露私钥。

A Go bug put the private key inside the certificate, in clear text (CVE-2025-6224). The culprit: h.Sum(key) doesn't hash key. It appends the hash of nothing to key. https://t.co/glhBARWG9Q https://t.co/sDJQ7Vwivn

cvecrypto代码审计 @PentesterLab source ↗

自动化扫描加 SSRF、任意文件上传已实战化,暴露面即入口。

RT @MsftSecIntel: Threat actors’ abuse of ARTEX continues to expand, with Microsoft Threat Intelligence observing financially motivated Storm-3358 using the agentic AI offensive framework alongside conventional automated tools to target financial organizations. https://t.co/SJn0rBvsU7 Storm-3358 has used ARTEX to scan internet-exposed systems, exploit security flaws including anonymous file upload and server-side request forgery (SSRF), and install open-source remote access trojans (RATs) for persistence and lateral movement. This activity illustrates a continuing trend of threat actors using agentic AI penetration testing frameworks to accelerate attack chains. Read our latest blog to get indicators of compromise (IOCs), along with mitigation, detection, and hunting guidance.

ai_agentssrf事件 @nas_bench source ↗

头注入的实际危害被低估,演示给出 CDN 内凭据窃取的可复现路径

RT @t0xodile: How bad is header injection really? Watch @m4st3rspl1nt3r and I weaponize it to create Desync worms, capture credentials inside a CDN and inadvertently share our shopping cart with thousands of users. The recording for CRLF-Powered Desync Attacks is live! https://t.co/zTKeCXrYJX

http_desync缓存投毒web @0xocdsec source ↗

暴露在公网的 N-central 应立即下线,攻击路径已被野外验证。

🚨 During the past 24 hours, we have observed probing of N-able N-central for the 0-days that were dropped publicly yesterday. We have seen opening of anonymous agent sessions, pulling appliance config and hitting the remote control endpoint without auth. These pathways are part of the 0-day exploit surface. If you run N-central exposed, take it off the internet now. Track exploit intelligence 👉 https://t.co/juLcSVcx7Q

0day未授权访问事件 @DefusedCyber source ↗

形式化验证落到广泛部署的生产代码,为 AI 辅助代码审计提供可复用路径。

RT @pushmeet: Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified – by any method, with or without AI – and at this depth. More than a research breakthrough, it’s a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats. Formal verification offers a higher standard of security; however, it’s extremely complex and time consuming, which is why in many cases it’s considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month’s time. Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices.  This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work. Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work. On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary. I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger,  Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership. https://t.co/kN1eRYQ8fF

形式化验证llm报告 @moyix source ↗

OAuth 同意滥用的下一步,AI 代理身份成为新入口,代理授权需重审。

RT @threatinsight: Proofpoint researchers identified BluePhishing, a potential attack technique that could be considered the next evolution of OAuth consent abuse. ⚠️ Attackers can use Microsoft Entra’s agent identity model to gain initial access to corporate environments. https://t.co/eioFaYsJpH

ai_agent绕过云 @virusbtn source ↗

构建链被污染且经苹果公证,签名公证无法再作为供应链可信凭据

Socket traced a backdoored bee-js package and an Apple-notarised Swarm Desktop update to Ethersphere’s own release pipeline. The backdoor got its C2 address from the same Ethereum contract as used in the earlier SleepyDuck campaign. https://t.co/ZFhZvJiPo3 https://t.co/xgjqy97kIM

供应链后门npm @virusbtn source ↗
Recommended15

自建 agent 与 C2 集成的真实取舍,动手前可避开典型设计陷阱。

I get the appeal of putting an agent harness in your C2 interface, but I don't think it makes sense. It (usually) means I can't reuse my ChatGPT/Claude subscription, and an agent that runs on the server won't have the tools I'd run over SOCKS or the local BOFs I want it to extend/debug. Beyond this, building a good harness is difficult. Just look at the Codex or OMP repos... many decisions impact model performance (prompt caching, compaction, tool output truncation, etc.).

ai_agentllm工具 @kyleavery source ↗

开源硬件即可发起 deauth 与 SAE 攻击,门槛降到一条命令。

esp-idf-injection-ng — Raw 802.11 frame injection on ESP-IDF v5+/v6+ via a one-command objcopy --weaken-symbol patch. Enables Authentication (SAE Commit), Deauth, Reassoc, Disassoc, and Control frames (RTS/CTS/ACK) through esp_wifi_80211_tx() https://t.co/i9AiJa4T8M https://t.co/7tEzlT4sqv

工具固件绕过 @KitPloit source ↗

给出该 PowerShell 规避手法的有效性与检测代价实测结论。

Doing some late-night testing with ScriptBlock Smuggling. In a nutshell, its key use cases include: ✅ Potentially evade AMSI Inspection ✅ Conceal malicious PowerShell activity ❌ Low Sophistication ❌ Noisy - Multiple Telemetry Events https://t.co/cGIy7yqKUo

amsipowershell绕过 @ipurple source ↗

签名伪造加 BITS 持久化成本低且隐蔽,附哈希与域名可立即布检测。

Backdoor leveraging BITS notify command for persistence. Signed "Barbaros Cetinkaya" adf0957559b8914b125ef263959f2882481df3b5dd0472888bf07e4e455a5d34 h/t @malwrhunterteam healthlys[.]com Also some sync-endpoint[.]azurewebsites[.]net domains https://t.co/XBIG7ATTci

持久化windows签名伪造事件 @SquiblydooBlog source ↗

恶意软件开始内嵌 LLM 助手,API 密钥查找与提示构造成为新检测面。

RT @joe4security: 🚨 NETRYX RAT Sample Reveals LLM Integration Under Development 🤖 🔥 A developing NETRYX RAT sample reveals an emerging LLM-powered operator assistant alongside remote command execution and persistence capabilities. ⚡ The RAT integrates prompt handling, API-key lookup, configurable AI models (default: gpt-4o-mini), and LLM request construction. 🔬 The AI integration is still under development. Missing API credentials prevented confirmed LLM queries, while incomplete response handling suggests unfinished functionality. ⚠️ The threat is real. The AI capability is emerging. No LLM-driven attacks were observed. 🔗 Full Joe Reverser Analysis Report: https://t.co/mCY0RhW4gQ #NETRYX #Malware #AI #ThreatIntel

llmai_agent恶意软件 @virusbtn source ↗

云上闲置的跨账户集成角色是易被忽略的横向通道,已在真实入侵中被利用

Wiz found an attacker using Qubole’s AWS credentials to get into customer accounts through integration roles that otherwise sat unused. In some cases, the attacker downloaded an entire S3 bucket in one session. https://t.co/To0P9hq8DO https://t.co/bxtaqbpfzt

云aws横向移动事件 @virusbtn source ↗

Agent 出网收敛是目前少数可落地的 Agent 运行时管控手段

RT @Cyb3rMik3: The recording from my session at CollabDays Greece is now live. I walk through how to combine Global Secure Access with Copilot Studio agents to control and secure agent egress traffic, so your agents only reach what you’ve explicitly allowed. Watch it here: https://t.co/AwtpoY4F7g

ai_agent网络议题 @ipurple source ↗
More47

一年期跟踪的 APT 工具链细节,含 rootkit 与电信目标 TTP。

RT @_CPResearch_: We spent a year tracking CopperDragon’s Sparrow Framework - used by Chinese nexus APTs around the 🌍 🔎 Watering-hole attacks on government websites 🕵️ Passive backdoors and rootkit capabilities 📡 Espionage targeting major telecom operators Read More -> https://t.co/oWeio9fZVa

aptrootkit报告 @_winterknife_ source ↗

本地权重推理并经雷雳 5 组网,影响自托管 AI 的部署选型。

RT @aaronp613: Apple is working on a new app called Harbor Harbor turns your Mac into a private AI inference server. Run open-weight models locally on Apple silicon—your data never leaves your hardware, and nothing reaches the cloud. Connect multiple Mac computers over Thunderbolt 5 to pool unified memory, enabling large models and higher concurrency that a single computer can handle alone. Set up in minutes, not hours Harbor automatically detects your devices over Thunderbolt 5 to form a cluster. Plug in, start, and you’re running—no manual configuration, no command line, no infrastructure expertise required. Download and run models directly Browse, download, and load open-weight models from within the app. Harbor handles model management so you can focus on the work, not the setup. Cluster for scale Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource. Run models that exceed any single computer’s capacity, or serve more concurrent requests across your cluster. Harbor is currently in a private preview under the Claris brand.

llm本地部署apple @0xocdsec source ↗

LLM 产品宣称的数据保留承诺与实际泄露面存在落差。

So, they laughed at people calling it a notetaking app despite supporting it from day one, forced us all into their LLM slop, jacked up token cost, and now our chats have been leaked even though ZDR… wake me up for the class action but meanwhile, Fuck You @NotionHQ https://t.co/LRyxWwY746

llm数据泄露事件 @_xpn_ source ↗

浏览器端载荷投递手法的完整拆解,可直接对照现有防护。

HTML Smuggling New Medium post, in this one we will see how HTML Smuggling works, a technique that lets you deliver a payload to a target machine without ever sending a file over the wire https://t.co/UQFBRquL2D https://t.co/cXaz3EUXwW

钓鱼浏览器初始访问 @papadelospapas_ source ↗

RT @ClaudeDevs: We've given all Pro, Max, Ultra, and Team users one free full reset, across all models and limits. Use it anytime before December 10 at 2:00PM PT. We recently fixed a few small issues that used more limits than intended, and the Fable resets didn't do much for those of you who'd already hit your weekly limits for all models. While these issues didn't affect everyone, we want to make up for the trouble. You'll find it in Settings > Usage, or at https://t.co/5YRzheAZyQ.

llm @0xocdsec source ↗

多 Agent 协作已落到漏洞挖掘与代码审查,改变攻防效率基线。

Spent last few weeks trying to coordinate swarms in a way that agents not only can interact with each other but share progress in real time and help improve each other, this has been v promising results in multiple areas, including research, vuln discovery, finding optimal solution, code review etc.

ai_agentllm漏洞挖掘 @0xAsm0d3us source ↗

LLM 能写底层驱动,网页侧硬件访问与驱动代码的可信边界值得留意。

RT @AliceInDisarray: Okay so lets get a few facts out of the way. macOS does not support LTE modems Apparently *webusb* supports LTE modems Claude is smart enough to write an mbim driver and tcp stack this website can talk to LTE modems https://t.co/4pPCSfCNCB

llm浏览器驱动 @0xocdsec source ↗

TLS certificates for https://t.co/1Ce79EfjpV and https://t.co/egUzEUTeAJ will be updated by Nov. 13. Afterward, you will need to re-download the latest Cobalt Strike updater (20261109) if you want to update Cobalt Strike. Cobalt 4.13 users need no action. https://t.co/cWS97XGjcm https://t.co/rdb5QwIb7C

工具基础设施 @_CobaltStrike source ↗

固件类报告的受理边界曝光,影响硬件 bounty 的提交策略预期

RT @0xrootRE: 特斯拉选择的漏洞审核平台 Bugcrowd 上面的审核员好像更倾向于或者说更擅长审核web平台API的安全漏洞,对于固件方面的安全问题并没有足够的耐心去审核(不排除是他们手上没有固件导致无法审核)纯文本的漏洞报告直接关闭 。已经没有足够的耐心和他们扯皮,也没有更多提交漏洞报告的动力和想法了。

固件bug_bounty @0xocdsec source ↗

白名单机制的实际盲区清单,来自 LOLBAS 维护者的一手经验

RT @magicswordio: One of LOLBAS founding contributors, Jimmy Bayne (@bohops) joins us for the next PLL Episode. If you’ve ever wondered what’s still executing after you think you’ve locked everything down, he’s the person to ask. We’re getting into WDAC bypass, the binaries that survive app control, and what attackers are still finding room to work with. Subscribe here to join: https://t.co/yL7QoxbpP0

wdac绕过lolbas @bohops source ↗

ARM 设备跨平台启动的可行案例,涉及固件与引导层改造

RT @qwerty__cs: 这不是 MacBook,是小米平板。 开发者 UniXNU 在小米平板 6S Pro 上跑起了 macOS 27.3,触屏能用,他说几乎所有功能都正常。 一块骁龙芯片的安卓平板,配上键盘套,开机就是苹果桌面。 https://t.co/i1YooBo4oX

固件移动端 @0xocdsec source ↗

Working on this crazy idea: reviving an 8088 PC with a custom OS. Boots from 5.25" floppy, FAT12/16, preemptive multitasking at 18.2 Hz, BASIC, dual consoles (Hercules mono + serial), 3270-style panel. No DOS. Huge shoutout to DeepSeek ;-) #retrocomputing https://t.co/TxhBS3gAu5

retrocomputingos @decoder_it source ↗

SEO 投毒分发、载荷抓浏览器凭据与钱包,Node.js 变种改造成本极低。

RT @Unit42_Intel: Unit 42 observed search engine results leading to a site impersonating RØDECaster that distributed #SantaStealer malware. The Node.js stealer harvests browser credentials, Discord tokens and cryptocurrency wallet info. Details at: https://t.co/9Nq2NKGcxS https://t.co/4wQr1bZXdB

恶意软件浏览器事件 @virusbtn source ↗

半年手法与新后门汇总,可直接转成狩猎假设与检测规则。

ESET has published its Q2-Q3 2026 APT Activity Report, which provides an overview of notable activities conducted by selected advanced persistent threat (APT) groups and investigated by ESET researchers between April and September 2026. https://t.co/98Y2GXZMa9 https://t.co/m7j8GzVmKF

apt报告后门 @virusbtn source ↗

多智能体自动攻防已进入实用阶段,攻击侧自动化节奏明显加快

RT @ngnicky: 최근에 ARTEX 기반의 Antifox AI 가 많이 사용된다고 합니다. 모니터링 강화!! AntiFox는 AI 기반 자율 침투 테스트 플랫폼입니다. 승인된 대상이 주어지면 플래너/워커/마스터로 구성된 멀티 에이전트가 정찰 → 공격 표면 발견 → 익스플로잇 검증 → 취약점 로깅 → 보고서 전달 과정을 이중 그래프 아키텍처 (전역 자산 그래프 + 작업별 탐색 그래프)를 기반으로 자율적으로 수행하며 , 완벽한 가시성, 개입 기능 및 전달력을 제공합니다. https://t.co/r69rOPu9Jh

ai_agent工具自动化 @artem_i_baranov source ↗

机器人批量造势推高应用声量,是识别协同虚假推广的可观察样本

If you look up “Gamma PowerPoint” on X, you’ll see hundreds of posts from Turkish bot accounts claiming that it’s the most amazing application of all time In reality, once you pay $216, it spits out the most horrendous looking presentations that look like they were made by GPT-3 Your standard Claude/GPT subscription does 1000x better They’re worth $2.1B btw We live in strange times

bots威胁情报事件 @C2IRIS source ↗