⚡Megatron LLM Hub
All digests
Twitter

推特安全流

Top7

无需修改内存保护属性就能执行载荷,可规避常见注入检测规则

- Scans for existing handles with sufficient access rights - Finds pre existing RWX memory regions without modifying protections - Writes shellcode directly into the selected RWX region - Queues a special user APC for immediate execution - (Optional)Encrypts shellcode using Speck https://t.co/PqyDCivk2q

进程注入绕过poc @papadelospapas_ source ↗

Agent 处理不可信目标输入即受影响,附统计数据与 TokenExhaust、HoneyAI 防守思路。

RT @JulioUrena: I spent weeks researching AI Pentesting Agents: can they be hacked? Yes. I got RCE on most Pentesting Agents ARTEX, strix, PentestGPT, etc. My research, stats, recommendations + 2 new concepts for Blue Teamers: TokenExhaust & HoneyAI are here: https://t.co/XyvJ369OkG

ai_agentrce报告 @0xocdsec source ↗

LLM 代理驱动的入侵已完成实战化,同类自建平台需按同一威胁模型加固

RT @FBICyberBranch: Threat actors are weaponizing ARTEX, an open-source, large language model (LLM)-driven agentic platform to automate multiple phases of cyber intrusions at machine speed. Since at least August 2026, a variety of malicious cyber actors have used ARTEX to conduct campaigns targeting organizations globally across multiple sectors, including financial services, government, and technology. ARTEX significantly reduces the time required to chain attack techniques and enables threat actors with minimal technical expertise to conduct complex, multi-target operations simultaneously. Today, the FBI released a FLASH highlighting the tactics, techniques, and procedures and indicators of compromise associated with the ARTEX framework. Organizations should follow the mitigation steps outlined in the FLASH to improve their security posture in response to ARTEX and similar agentic SI frameworks: https://t.co/l9lgbqWNYk

ai_agentllm事件 @artem_i_baranov source ↗

现成检测规则可直接部署,覆盖 ARTEX 及其 AWS 分支 AntiFox 的具体行为。

RT @malwaremustdie: #ARTEX signature detection pack #updates v1.1.0 is out (#MMD-072-2026), built from running own #Playbook (works well!) w/more tests against recent network traffic dataset, have improvements summarized below: • Rename-proof webshell rules (upload → execute, any filename) • Windows coverage: IIS/PHP workers spawning shells, script drops • AWS rules for AntiFox, an ARTEX fork (secrets/IAM/S3 sweeps) • OAuth high-privilege consent + Graph recon (Sentinel) • LLM-API egress policy rules for servers • IP IOCs: 90-day alert window, then hunt-only • New durability labels D0/D1/D2 + mutation testing *) RC1 rules unchanged. Known gaps documented, not hidden. *) Read Blog(Appendix E), Changelog in v1.1.0 *) Remember #BehaviorFirst, #FightAIwithAI *) Thanks to friends who shared/sent their recent traffic, Links: Blog(Appendix E): https://t.co/xxOdSawqCz GitHub(same): https://t.co/Ep6kJjb7UO Comments(same thread): https://t.co/yFlOowD5OB #malwaremustdie - @unixfreaxjp

ai_agent工具云 @0xrb source ↗
Must-see9

自托管 LLM 应用可复用搜索、OCR、语音能力,降低搭建成本

RT @MaxForAI: 🚨DeepSeek居然把他们整个App也开源了。。。。 刚刚 @deepseek_ai 正式发布全家桶应用API,直接把DeepSeek App里几乎所有的功能都开放给开发者了。 包括联网搜索、深度思考搜索、文件解析、阅读链接、语音交互,以及OCR、ASR、TTS等底层能力。 甚至官方还直接开源了一套完整的聊天前端Demo,支持导入DeepSeek历史对话。 也就是说,现在开发者拿着DeepSeek的API,基本就能1:1复刻一个DeepSeek App。 DeepSeek在公告最后还专门表示:「DeepSeek现有的对话产品只是较为基础的一种形态。现在,官方产品背后的各项能力均已支持通过 DeepSeek API 随心调用、自由组合,我们期待开发者能够借此省去重复的基础工作,打造出更多样、更极致、更具想象力的产品。欢迎使用 DeepSeek API,尽情创造。」 现在各家AI公司都在拼命做自己的超级入口,想方设法把用户留在自家App里。 DeepSeek倒好,直接把自己App背后的能力全部开放出来,连复刻方案都给你准备好了。 你想做一个自己的DeepSeek?可以。 你想把DeepSeek的搜索、语音、文件处理能力拆出来,重新组合成一个新产品?也可以。 甚至可以在DeepSeek已有的能力之上,做出比官方App更复杂的产品。 这才是这次发布最值得关注的地方,DeepSeek对自己在AI生态里要扮演什么角色,似乎想得很清楚。 就是下次能不能别在北京时间的周五晚十点发布了。。。。

llmai_agent工具 @kernullist source ↗

进程注入是主流恶意软件的基础手法,本文对检测与绕过的细节值得参考

RT @SpiderMorpheus: Process Injection is one of the most fundamental capabilities of malware. It plays a role both in establishing a foothold after initial access and during post-exploitation, but the technique suffers from a fundamental problem: its traditional behavior is predictable and well understood by defensive solutions. The classic Remote Process Injection flow can be roughly summarized in four stages: 1. Enumerate processes on the compromised machine; 2. Obtain a handle to the target process; 3. Allocate and write shellcode into the target process's memory space; 4. Change the memory region's permissions from RW to RX. The set of WINAPIs designed for these operations is relatively small, which makes life easier for EDRs. The list of the most sensitive APIs is short: VirtualAllocEx → NtAllocateVirtualMemory VirtualProtectEx → NtProtectVirtualMemory WriteProcessMemory → NtWriteVirtualMemory Whether through kernel callbacks or userland hooks on ntdll, these calls are easily monitored. There are ways to deal with this problem, including simply not performing process injection, at least not during initial access, and choosing something like DLL Sideloading. But we want a challenge, right? :) In the screenshot I show something different. It is possible to write into a remote process's memory without invoking sensitive APIs such as NtAllocateVirtualMemory or NtWriteVirtualMemory. Based on an article written by the amazing researcher @TwoSevenOneT, I developed a shellcode runner that successfully evades Windows Defender (with Cloud-Delivered Protection enabled) by poisoning a target process hStdInput. The technique abuses Windows console applications (such as netsh.exe) by redirecting the child process's hStdInput handle via CreateProcess with STARTF_USESTDHANDLES. The content written to the Named Pipe ends up allocated in the target process's memory by the operating system itself, with no explicit call to VirtualAllocEx or WriteProcessMemory. To locate the shellcode in remote memory, the payload is preceded by a previously established byte sequence that serves as a marker. The runner scans the target process's memory for this pattern, identifies the region, adjusts permissions, and redirects execution to the shellcode. None of the most sensitive APIs are used for writing the payload. Three APIs that are also used, among other purposes, in traditional injection are present (VirtualProtectEx, CreateProcessW, and ResumeThread), but outside their traditional context they are far less suspicious. Currently, this shellcode runner PoC would not successfully evade top-tier EDRs, but the goal was to demonstrate viability, not perfection. A little bit more evasion tomfoolery would be enough. The link to the article that guided me through this experiment will be down below, make sure to do something useful with your life and read it. Keep Hacking, Sp1d3rM_*!^

进程注入绕过报告 @0xocdsec source ↗

AI 编码基础设施被逐层穿透,凭据与插件分发构成关键攻击面。

RT @hrkrshnn: This is a crazy report. 1) Fireworks breach 2) Traces from Cursor users exposed 3) Traces had customer credentials 4) Cursor credentials were also exposed 5) Attacker replaced Claude Code inside Cursor marketplace with the cursor credentials. The attackers are really after AI infrastructure. Ruthless.

供应链ai_agent事件 @cyb3rops source ↗

宏、ISO、LNK 被堵后 MSC 成为新入口,漏洞八年未补,检测思路可直接借鉴。

RT @0xpwnie: Microsoft blocked Office macros in 2022. Attackers moved to ISO files. Microsoft fixed ISO files. Attackers moved to LNK and OneNote. Microsoft closed those too. By mid-2024 attackers had landed on MSC files, the administrative console files used by IT teams to manage Windows systems. GrimResource exploits a cross-site scripting vulnerability in apds.dll that was reported in 2018 and never patched. The sample that exposed the technique was a red team payload accidentally uploaded to VirusTotal by a blue team analyst who picked it up during an assessment. When it first landed on VirusTotal on June 6 2024, zero antivirus engines flagged it.

绕过钓鱼poc @0xpwnie source ↗

AI 辅助入侵下单条告警失效,关联检测成为阻断关键。

RT @SBousseaden: AI-assisted intrusions aren't just about speed. In this CyberStrikeAI case, 78 low/medium severity AWS alerts across 16 days highlight why high-signal correlation matter: connecting weaker signals to detect and disrupt an intrusion before it progresses. https://t.co/KJHGDpJjtm

ai_agent云事件 @cyb3rops source ↗
Recommended15

Linux、Android 上开箱即用,可排查 Agent 出网流量或做取证。

ecapture v2.11.0 — Capture SSL/TLS plaintext with eBPF—no MITM proxy or custom CA installation. Supports Linux and Android on x86_64\arm64 and loong64. https://t.co/RHoj7fMILf https://t.co/AFF4prCPUh

工具取证tls @KitPloit source ↗

嵌入式 Web 服务器装量巨大,三处 TLS 缺陷可直接影响服务端加密信任

''Choose Your Backend: Three TLS Bugs in GoAhead - FoxIO | Cyber Innovation'' #infosec #pentest #redteam #blueteam https://t.co/WA5wRDjOaa

tls漏洞固件 @CyberWarship source ↗

NAS 常暴露公网,MCP 接口让 agent 直接执行审计与修复

synology-nas-cli — Zero-dependency Python CLI and MCP server for Synology DSM 7 that audits NAS configuration against the C3RB3R/.L0CK3D ransomware campaign, plans fixes, and applies them. https://t.co/nTebkPQkCC https://t.co/5GZ02CYL1l

工具ai_agent加固 @KitPloit source ↗

无需密钥即可跑完整调查流程,明显降低 OSINT 上手门槛

OsintHunter — 🎯 The ultimate OSINT toolkit: investigation graph, confidence scoring, alias families, map, timeline and reports. 49 modules, 400+ sites, no API keys required. https://t.co/yCYHjHcleQ https://t.co/R7o1KG4qZl

工具osint情报 @KitPloit source ↗

把投入从「挡住初始访问」转向遏制,可压低入侵后的爆炸半径。

RT @dinodaizovi: This is the thing that I think organizations should be *most* focused on. You all run a ton of 3rd-party code that is mostly outside of your control, but here is what is in your control: containing and slowing down compromises *after* initial access to remain a size that you can manage. Your number one goal should be eliminating possibilities for lateral movement and global privilege escalation (i.e. different than LPEs).

防御提权 @S1r1u5_ source ↗

引用真实故障编号显著提高可信度,故障后 24 小时内为钓鱼高发窗口。

RT @MsftSecIntel: Microsoft Threat Intelligence observed an active phishing campaign exploiting the Microsoft 365 service disruption on November 12. The messages impersonate Microsoft 365 service health notifications and reference the legitimate Exchange Online incident EX1550543. Some of the phishing emails Microsoft detected on November 13 falsely claim that messages were held during the outage and will be deleted within 24 hours unless recipients click to restore them. Observed subjects follow the pattern "Outage - [six-character code]," with sender display names using organization-specific prefixes followed by "Outage Notification" or "Service Health." The campaign uses three sender domains, moments-notice[.]com, planetmaison[.]com, and newbooker[.]com, with recipient-specific SendGrid click-tracking links redirecting to reallifeaccounting[.]com, including the observed path /themelist0.html. Similar messages targeted organizations across multiple US sectors. The activity is consistent with credential phishing, although credential collection and malware delivery have not been confirmed. Security teams can use Microsoft Defender for Office 365 Threat Explorer and Microsoft Defender XDR advanced hunting (EmailEvents, EmailUrlInfo, and UrlClickEvents) to identify suspicious messages, examine embedded URLs, and investigate user clicks. Microsoft Defender for Office 365 Safe Links provides time-of-click protection against malicious URLs when properly configured. Organizations should verify their Safe Links policies, investigate potential user interactions, and confirm service disruptions directly through the Microsoft 365 admin center.

钓鱼云事件 @0x534c source ↗
More28

针对 Apple 为高风险用户提供的防护做跨版本实测,可用来检验现有加固措施

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

移动端隐私议题 @0xocdsec source ↗

官方通报的在野手法,社工链条完整,可据此调整验证流程。

In the 'FaceStealer' campaign, North Korean operators combine pre-recorded webcam footage of previous victims with AI deepfakes of industry figures to make fake video calls look authentic. Rather than engaging in live conversation where a target might ask verification questions, the hackers immediately simulate a broken microphone. The call appears continuous for just a few seconds; long enough to build visual trust before a chat prompt tricks the victim into installing malware to "fix" the audio. https://t.co/WiWvbuMvoq

钓鱼社工事件 @0xpwnie source ↗

RT @OpenBinAI: OpenBin update! 🔧 • Ghidra-style function tree: Functions, Imports, Thunks, auto-grouped by prefix • Data XREFs: click through to every function that references a symbol • Syntax-colored, clickable Program Listing • Smarter pseudocode colors https://t.co/Sbxod6E6Wa https://t.co/RBm4XGyPap

工具逆向 @0xpwnie source ↗

说明内置编码器已被签名后,最小代价的自建混淆改造路径。

RT @SpiderMorpheus: Funny enough, basically the only three things I did to the original PoC was: - Writing it in C, because we love C - Adding Dynamic API Resolution for some APIs (honestly, for defender I feel like this has been an overkill for this specific technique but I am working on a complete version so I threw it there) - Meterpreter shellcode was encoded using IPv6fuscation instead of one of the baked-in encoders because the encoders themselves are signatured as fuck

绕过pocshellcode @0xpwnie source ↗

uBlock v1.76.3b0 — Wide-spectrum content blocker for browsers that blocks ads, trackers, coin miners, and malicious sites using filter lists and customizable privacy rules. https://t.co/KkKaYznks0 https://t.co/OdNi84pzzP

浏览器工具 @KitPloit source ↗