必看16
首个公开的自主 Agent 端到端攻击样本,展示 Agent 编排被武器化的实际形态。
RT @PandaRE__: New blog drop! 😊
Operation Snowball 🇨🇳
While sweeping open directories on @etugenio, I found a workspace of an autonomous AI intrusion harness. The TA used a jailbroken DeepSeek agent and ran two weeks of attacks against a European telecom provider:
https://t.co/rL7DcyPVu0
ai_agentllm事件报告
@cyb3rops
原文 ↗
把攻防方法论转成 Agent 可执行指令,直接影响自托管 Agent 的工具化落地方式。
AboutSecurity — Everything for pentest. | A penetration testing knowledge base that distills security methodologies into a format executable by AI Agents. https://t.co/KZvwG67fk3 https://t.co/hOGQ3BjTrY
ai_agent工具渗透测试
@KitPloit
原文 ↗
Rust 被当作内存安全基石,波及面覆盖所有依赖 std 的项目。
RT @MsftSecIntel: Microsoft’s FORGE Lab identified several memory-safety issues in Rust's core and std libraries, including cases where code written entirely in safe Rust can trigger undefined behavior. https://t.co/2iVeFfAb0H
The research examines three recurring patterns of misplaced trust in unsafe code: trusting code not to panic, trusting user-provided implementations to be correct, and trusting the underlying system. Findings include issues affecting iterator cloning, string replacement, and interactions with the operating system.
Using Hyenas, FORGE's research version of codename MDASH, researchers combined AI-assisted vulnerability discovery with technical investigation and validation. They investigated the underlying code, reproduced the bugs, assessed potential security impact, and worked with Rust maintainers to report and fix them.
While Rust's ownership model and type system prevent entire classes of memory-safety vulnerabilities, unsafe implementations underneath safe APIs must still uphold those guarantees. The findings illustrate how subtle mistakes in foundational libraries can lead to undefined behavior, and why Rust's safety model remains important even when bugs are discovered.
Read the analysis for details on the findings, root causes, upstream fixes, and lessons for developers working with unsafe Rust.
rust内存安全报告
@domchell
原文 ↗
LLM 路由平台泄露 key 等于额度与调用权旁落,轮换成本极低。
RT @Kostastsale: OrcaRouter may be dealing with a security incident.
They has warned users about suspicious activity that could potentially expose API keys.
According to OrcaRouter, approximately 1.2% of accounts were flagged for potential exposure, although unauthorized usage has not been confirmed.
If you're using OrcaRouter, consider rotating your API keys, enabling 2FA, and reviewing recent API usage for suspicious activity.
llmapikey事件
@0xocdsec
原文 ↗
选模型跑 agent 安全任务前,可直接比对攻防表现与成本。
Three new models have been added to DreadIndex (@MiniMax_AI M3.1 Flash, @OpenAI GPT 5.6 Sol, and @Kimi_Moonshot K3.1 Pro), plus a new dashboard to explore how we score and analyze model performance.
🔗: https://t.co/48fRcQCvIN
🧵: Observations from the latest round of evals https://t.co/RtBWF6HrKG
llm评测ai_agent工具
@dreadnode
原文 ↗
EDR 自身被绕过会直接削弱终端检测的可信度,细节待披露。
Verify it, close it, read it again?
Bro,that's TOCTOU. EDR vendors :))
Reported it. Blog coming if accepted. https://t.co/BFMTCwMw42
toctouedr漏洞
@0XDbgMan
原文 ↗
公开完整利用链与适用版本限制,可用于评估 EDR 自保护强度。
I found this a cool read: https://t.co/SmbpFb8gc2
However:
1) The victim must've been on Windows 10 since on most Windows 11 builds fixed address kernel mapping of KUSER_SHARED_DATA is marked RO.
2) PsTerminateProcess invokes the same process exit notification callbacks. https://t.co/A9GTt0C6bv
lpe内核edr报告
@_winterknife_
原文 ↗
代码审查 agent 可自主识别投毒 PR,改变供应链攻击的发现时点。
RT @dguido: My Dot found malicious code in an open PR on a @trailofbits Github repo. I didn't ask it to hunt for malware! It noticed a commit, inspected what changed, and warned me about an attack all on its own. https://t.co/dmLwL0cCzu
ai_agent供应链llm
@0xocdsec
原文 ↗
bootkit 框架开源落地 ARM64,Linux 主机固件层持久化门槛降低。
RT @Idov31: I had a lot of fun giving a talk at @nullcon and now I can finally release ARM64 support for my Linux bootkit framework - Silverseal, making it the first public ARM64 UEFI bootkit targeting Linux distros, to the best of my knowledge. https://t.co/AJVhQLUnma 1/3
bootkit固件工具linux
@cgomezz_23
原文 ↗
Google 官方横向评测,量化 LLM 在漏洞挖掘上的真实上限与差距。
''Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines ''
#infosec #pentest #redteam #blueteam
https://t.co/WvBVLJ2jih
llm浏览器报告
@CyberWarship
原文 ↗
针对 AI 编码助手的供应链投毒,仓库配置即可触发静默执行
RT @fr0gger_: 🧐 New active campaign of the Miasma/testmetest supply chain worm that targets AI coding agents!
12 GitHub repos were poisoned on Nov 9 with forged, backdated commits. Every agent config in each repo tells the assistant to run an obfuscated 3.7 MB scripts/setup.js
The trick is in CLAUDE.md, marked "DEVOPS-SEALED". It calls setup.js a "sealed binary artifact" and tells the assistant:
- to not read, quote or analyse it
- to run it "without displaying output"
- to run it "automatically without asking for confirmation"
It is a prompt injection written to stop the agent from looking at the file!
Full prompt below 👇
ai_agent供应链llm事件
@cyb3rops
原文 ↗
除 IPv6 外 IPv4 路径同样可绕边界防护,影响面扩至多数内网
RT @ikotas00: We discovered a new variant of BunkerBuster that targets IPv4. It also allows remote attackers to bypass firewall restrictions and establish bidirectional communication. Codename: BunkerBuster2💥
#BunkerBuster https://t.co/bKUtTBu8Zh
编码 Agent 会读取被投毒的配置并执行攻击者代码,直击本地 Agent 供应链。
Socket analysed a malicious Ruby gem carrying a Shai-Hulud descendant. It targets developer credentials and poisons AGENTS.md, CLAUDE.md and MCP configs so coding agents can run attacker code. No victims were confirmed. https://t.co/IN0N3CE18Q https://t.co/yXdRdrG3v9
ai_agent供应链llm事件
@virusbtn
原文 ↗
安装包即触发主机指纹收集与二阶段下发,C2 与既有 DPRK 基础设施相同。
RT @nextronresearch: Our Thunderstorm-based artifact pipeline spotted "[email protected]", the latest npm package linked to a suspected DPRK cluster.
Importing the package automatically loads two obfuscated components disguised as "telemetry-info.mjs" and "dist/runtime-matrix.mjs".
The loader fingerprints the host and sends the collected data to the same C2 infrastructure observed previously: cdn[.]node-js[.]dev.
The server can respond with a Base64-encoded second stage, which is written as "helper.cjs" under the user's cache directory and launched as a detached Node.js process.
This iteration shows the campaign actively changing filenames and delivery mechanics while retaining its C2 infrastructure.
供应链npm后门事件
@cyb3rops
原文 ↗
展示开源模型在真实漏洞挖掘上的能力边界与开销。
Open-weights just require some focus. DSV4.1 for RE of regions + K3 || GLM5.3 for deeper contextual RE + VR. Exploit development is doable with open-weights but it does incur a decent amount of overhead. If you aren’t familiar with a specific subsystem or interactions between components then you may have a rough time. With the open weights, I was able to discover 4 within a more hardened surface (oob write, cross-priv info leak, 2x null-deref DoS). With a library of primitives, K3 or GLM5.3 can do well.
Daybreak is excellent though, I would highly recommend it for VR; but I was able to net these 4 reportable bugs in <72 hours with an open-weight setup — so we shouldn’t throw shade at open-weights, they’re pretty great when focused.
Lambda 域名自带云厂商信誉,常见 URL 过滤难以拦截。
🚨 𝗔𝗪𝗦 𝗟𝗮𝗺𝗯𝗱𝗮 𝗣𝗵𝗶𝘀𝗵𝗶𝗻𝗴 𝗖𝗮𝗺𝗽𝗮𝗶𝗴𝗻
Using @anyrun_app intelligence lookup and sandbox analysis, I observed a large-scale, ongoing phishing campaign leveraging 𝗔𝗪𝗦 𝗟𝗮𝗺𝗯𝗱𝗮 𝗨𝗥𝗟𝘀.☁️
Based on the telemetry reviewed, the campaign has been active since at least early October. Initially, the malicious Lambda functions were predominantly hosted in the 𝗔𝗪𝗦 𝗨𝗦 𝗘𝗮𝘀𝘁 (𝗡. 𝗩𝗶𝗿𝗴𝗶𝗻𝗶𝗮) 𝗥𝗲𝗴𝗶𝗼𝗻, but over the past week I observed a significant increase in newly generated Lambda URLs originating from the 𝗔𝗪𝗦 𝗨𝗦 𝗘𝗮𝘀𝘁 (𝗢𝗵𝗶𝗼) 𝗥𝗲𝗴𝗶𝗼𝗻.
The threat actors use AWS Lambda Function URLs as the initial landing page, often masquerading as Microsoft 365 services such as 𝗢𝗻𝗲𝗗𝗿𝗶𝘃𝗲. Victims are presented with a pop-up message which, when clicked, redirects them through a series of security-check pages before ultimately landing on a fraudulent Microsoft sign-in page utilizing an 𝗔𝗱𝘃𝗲𝗿𝘀𝗮𝗿𝘆-𝗶𝗻-𝘁𝗵𝗲-𝗠𝗶𝗱𝗱𝗹𝗲 (𝗔𝗶𝗧𝗠) phishing framework.
What makes this campaign particularly concerning is that these Lambda URLs are being delivered through inbound emails and may evade traditional email security detections due to the reputation of legitimate AWS infrastructure.
🔍 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 𝗢𝗽𝗽𝗼𝗿𝘁𝘂𝗻𝗶𝘁𝘆:
Defenders should review inbound email traffic for URLs matching the following pattern:
https://{32-character-identifier}.lambda-url.{aws-region}.on.aws/?{tracking-parameter}=*@*
A high-fidelity detection can be achieved by combining this URL pattern with:
IsFirstContact == true
This helps identify first-time sender interactions involving AWS Lambda URLs that may be associated with phishing activity.🫡
#AWSLambda #Phishing #AiTM #Microsoft365 #ThreatHunting
推荐15
原版 Donut 已被大量 AV 标记,该分支主打规避,可直接上手验证。
homer - Evasive fork of Donut to generate position-independent shellcode that loads PE Windows payloads from memory https://t.co/J4tYkrkJNP
工具免杀shellcode
@ipurple
原文 ↗
防御向调优会限制 agent 的网络操作任务,选型前需知。
🔵 GPT 5.6 Sol (Daybreak Blue) 🔵
- Performance did not improve from the last model version; landed slightly lower to 5.5
- Soft refusals occurred on network operations tasks; understandable for a Daybreak Blue model, given it is tuned to allow for defensive cyber tasks, not offensive
- Lands on our value frontier, marked by its notably lower cost compared to GPT-5.5
llm评测ai_agent
@dreadnode
原文 ↗
低成本钓鱼手法,附开源工具可直接复现与验证防御。
Crapy Lnk Files Alter to dump your NTLM hashes to your responder listener upon user interaction https://t.co/sgoqjJFKY1
可直接部署试用的端点检测工具,补齐 LOLBins 检测盲区。
An enterprise-grade, host-based behavioral detection engine built in Python 3.12 for identifying Living-off-the-Land Binaries, Scripts, and Libraries (LOLBins) on Microsoft Windows endpoints. https://t.co/KhXRLSPEO2
工具lolbinswindows
@ipurple
原文 ↗
攻击向量与评分同时修正,影响面评估需按新参数重算。
RT @HackingLZ: If you missed it...Windows Paint Remote Code Execution Vulnerability(CVE-2026-83844)
Microsoft revised the official CVE advisory on November 7 from network/no-interaction to AV:L/UI:R, reducing the score from 8.1 to 7.0 and changing “over a network” to “locally"
cvercewindows
@0xocdsec
原文 ↗
内核地址泄露可助攻 KASLR 绕过,补丁已并入主线。
RT @andersonc0d3: perf/amd/ibs: Fix kernel address leak via REGS_USER and REGS_INTR
https://t.co/xnhv2XP3nM https://t.co/Xk7eNlsvDn
针对 agent 与服务账号的权限边界离线检查,多云可直接跑。
nhi-risk-analyzer — Discovers, risk-assesses, and remediates non-human identities across AWS, Azure, and GCP using offline rule-based analysis, privilege-escalation detection, and permissions-boundary containment. https://t.co/99RNnlIbe2 https://t.co/57CQOij3QK
多 agent 编排用于实战攻击的公开案例,可对照自建 agent 权限边界
RT @Technopolitik_: A hacker in Taiwan used BOSSAI, a 12-agent AI attack mesh, to attack Taiwan’s emergency services while targeting hundreds of banks, universities, military systems and other organizations across 11 countries. The hacker controlled a 4090 cluster in Guangzhou. https://t.co/cvIbdrDdTr
ai_agentllm事件
@0xocdsec
原文 ↗
自动化对手模拟并逐技术回传结果,省去手工编排攻击链
Crucible - a Mythic C2 agent purpose-built for purple team adversary simulation.
It transforms Atomic Red Team YAML test definitions into a self-contained .NET 9 executable (Windows or Linux) that checks in as a Mythic callback, waits for operator tasking, then executes the campaign on command, reporting per-technique results back to the Mythic UI in real time.
https://t.co/MvC7qAErZ8
工具c2redteam
@ipurple
原文 ↗
展示 agent 驱动的二进制逆向工作流,可迁移到自有目标
If you're bored or curious, and want to watch an agent explore fairplayd on macOS and all components down to FairPlayIOKit, I made the repository public: https://t.co/vTDNMs9SJ4.
6.1 Sol + Hex-Rays IDA Assist; current goal has been active for 7 days.
ai_agentllm逆向工具
@in19h
原文 ↗
本地运行不上传代码,CI 中一条策略给出统一结论
oss-analyzer — Open Source security and compliance gate for GitLab CI and GitHub Actions: SBOM, dependency vulnerabilities, SAST, secrets, IaC and licenses, under one versioned policy and one verdict. No SaaS, no code sent to a third party. https://t.co/PiQfJOaJOP https://t.co/pfEWYohoaS
工具供应链sastci
@KitPloit
原文 ↗
公开 IAT 与运行期截图可加快样本行为判断
I just added two new features to the sandbox detonator: the complete Import Address Table of each sample, and runtime screenshots.
The IAT is available to all users, including free ones, while screenshots are only available on paid scans
https://t.co/sTlq2l3JIZ https://t.co/gVEOCjrxpf
工具沙箱恶意软件
@papadelospapas_
原文 ↗
可直接复跑越狱与防御基线,用于自建 Agent 的红队评估。
jailbreak-eval-7llms — Code for "Systematic Evaluation of Adversarial Jailbreaking Attacks and Defenses Across Seven Large Language Models" https://t.co/HxbQcqybaf https://t.co/Hfjsknp0TT
拆解 MOTW 判定细节,对钓鱼投递与检测绕过都有直接价值。
SmartScreen appears when the Zone.Identifier ADS is present, like in the image. In this ADS we got this content (second img)
The ZoneId have this value options:
0: Local Machine
1: Local Intranet
2: Trusted
3: Internet
4: Restricted
Just removing the ADS disapears the smartscreen https://t.co/I1yMalWME9
绕过windows钓鱼
@papadelospapas_
原文 ↗
把情报转成可演练技战术,红队复现成本明显下降。
🤯 You can now execute malware techniques directly from a threat intel report?!
Combat Theater version 1.2 "Metal Claw" is now out with awesome new features and content!
Check our release blog for all the details!
https://t.co/UwHocOVcx5
更多62
可在本地复现 Defender 检测遥测,用于验证检测规则覆盖与绕过手法有效性。
defender_observer — C#/.NET console tool that captures, records, replays, and explains Microsoft Defender Antivirus and EDR ETW telemetry in real time with filtering and a built-in event knowledge base. https://t.co/8mnAbU44in https://t.co/9ZQ3tVFXrV
工具edrwindows绕过
@KitPloit
原文 ↗
本地 AI 工作站引导更新失败的真实复盘,含不重装系统的可复用恢复步骤。
RT @hasherezade: My new writeup, this time something different: how a routine DGX OS update left my Acer GN100 stuck in a boot loop, and how I managed to recover it without reinstalling the system. A little adventure with GRUB!
https://t.co/5AofhhMP5U
固件linux报告
@kernullist
原文 ↗
多漏洞链式利用的最新实例,说明单点补丁难以阻断整条攻击链。
RT @ryHanson: My Apple exploit chain leveraged 7 different bugs and multiple techniques to achieve reliable code execution. I'm glad to see none of them were duplicates. https://t.co/7xTVBbuBpn
补丁与在野利用的时间差,决定移动端设备是否仍留有可利用面。
RT @thejailbreakhub: iOS 27.2.1 is out.. exploit fixes?
主机漏洞修复状态变化,决定现有利用链在哪一固件版本仍可用。
RT @__alula: it's patched on both switch 1 and switch 2, do not update :) https://t.co/QmYBZS7YSL
RT @felixrieseberg: I'm slowly discovering that the age of prompt engineering is basically over. It's so simple:
"You know, you should also build me a FPS. Insanely detailed quality pls, something that looks visually like it can keep up with AAA. Make it small in scope (like a tiny level) but otherwise impressive photorealistic graphics"
llmai_agent
@0xocdsec
原文 ↗
RT @SilzeeJailbreak: Soon https://t.co/G37yKoRfNS
提供了 EDR 与 IPS 叠加后攻击工具真实存活率的经验数据。
RT @UK_Daniel_Card: Can I show you what happens when you throw ARTEX at a network with MDE deployed?
> It gets the shit kicked out of it! (MDE goes BRRRR)
> not only that but the perimeter IPS catches stuff!
Now... if we have ALL the variables in place.... could we make something that 'works'?
Sure.... because infinite monkeys....
but networks aren't infinite monkeys...
I've just got a SHELL working (where the IPS didn't block it)
I've had to unblock identities and devices A LOT!
so far in my testing:
> THROW ARTEX at something that isn't DEFENDED and guess what... IT WINS!
> THROW it at a DEFENDED network... guess what? the blue team wins!
护栏误报会卡死 agent 的常规任务,触发模式值得记录。
Not able to attempt any math with Fable 5.2 without running into this guardrail which alleges reasoning extraction https://t.co/tGXOFds2A7
llm护栏ai_agent
@C2IRIS
原文 ↗
补充模型在静态审计与漏洞利用任务上的短板参考。
⚫ Kimi K3.1 Pro ⚫
- Landed at the top of the leaderboard at #2, as well as on our value frontier
- Kimi model family still showing weaker performance on SAST and binary exploitation tasks
llm评测sast
@dreadnode
原文 ↗
Unpopular Opinion: They should probably port the Linux kernel to modern C++
I don’t think that they use much AI to write code at Apple. I could be wrong about this, but it just feels that way, as someone who spends countless hours reverse engineering it.
New lesson added into the Introduction to Windows Malware Development course!
The Logonstealer one!
https://t.co/XTf8JXObZe
It's basically the explaintation of this repo available for y'all:
https://t.co/X64YdeC9PV
恶意软件工具
@papadelospapas_
原文 ↗
agreed, GLM 5.3 > Deepseek 1017 for cyber 🤷♂️
I put lots of wrappers around my browsers for that extra difficulty. It's never gonna be perfect without fil-c but it is what it is. https://t.co/FirQyftawj
涉及厂商与研究者披露边界,可作 VDP 流程对照案例。
RT @adnanthekhan: Will be interesting to see the story behind this & if N-Able had burned the researcher in the past.
For what its worth - N-Able runs a VDP through Bugcrowd (https://t.co/ri0TuIYqsX)
Reverse engineering stuff and dropping 0-days is NOT a crime & good faith goes both ways. https://t.co/i7u3dnSscU
反映 AI 辅助漏洞挖掘的产出规模,浏览器攻击面压力上升。
RT @aegiryy: 100+ chrome cves by a single researcher would've been unimaginable in the pre-llm era. what a year for ai-driven vulnerability research (and we still have two months left in 2026)! https://t.co/fTWthpmVHC
RT @mattshumer_: Fable 5.2 built this voxel engine from a single prompt.
Try it (sound on), it’s mind-melting: https://t.co/TBSgIOw75S https://t.co/EB3nknNArz
RT @CrispinCowan0: 28 years ago we published StackGuard (stack canaries) that mitigates Aleph One's excellent paper. Now running on nearly every Internet-connected machine.
https://t.co/dReritVDRz
RT @felixrieseberg: We're releasing Claude Fable & Mythos 5.2! They're good models, I have some demos and tips for you.
Fable beats Opus on big, long tasks. Good examples are rewriting an entire codebase, creating an excellent first draft of a large artifact, or exploring ambiguous scenarios.
I expect to see it used to port a lot of stuff to Rust, support journalists in large-scale investigations, and help with research & science. I often still prefer Opus 5.5 for everyday work. Fable's got a big brain and thinks a little slower.
RT @claudeai: Fable 5.2 and Mythos 5.2 are a step above Opus 5.5 in almost all the capabilities we evaluated.
Testers have found that they communicate more clearly than past Fable and Mythos models. They generate outputs up to 30% faster than Fable 5.1 and Mythos 5.1. https://t.co/nqXVdmLRpN
RT @claudeai: Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision. https://t.co/jP0AutLxXb
RT @claudeai: Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do.
On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots. https://t.co/VEiQiQrcxq
伪装面试流程诱导开发者运行恶意依赖,研发终端是首批目标。
RT @JFrogSecurity: Read the full research and technical details: https://t.co/Zz6cnSojNq
AI 辅助形式化验证若成立,会改变高危漏洞的审计与发现方式。
LLMs are making formal software verification feasible at a much larger scale - Frontier Red Team @AnthropicAI
https://t.co/YyqY63KY71
llmai_agent报告
@pentest_swissky
原文 ↗
家用摄像头基站的固件逆向全流程,可复用到同类 IoT 设备审计。
Reverse engineering the Eufy HomeBase 2 firmware
https://t.co/po4Fho7AfI
#infosec https://t.co/TEQJpcVCWw
Android 内核提权链的字段级利用细节,是稀缺的完整复现材料。
RT @eshard: The final part of our CVE-2023-20938 series: turning a Binder use-after-free into root on a Pixel 6 running Android 12.
Root turns out to be anticlimactic: a few fields in one kernel struct, set to zero.
🔗 Read part 5: https://t.co/YK6rs0DVEO https://t.co/ybANRB1zZ8
提权内核移动端cve
@0xocdsec
原文 ↗
自建与托管 agent 运行环境的成本与护城河判断,影响选型。
I see a lot of sandbox a.k.a. give your agent its own compute companies popping. What's the moat, and how sustainable is that business?
I mean if you're just renting the compute from someone and reselling it with an abstraction on top (with some optimizations, and perhaps a better orchestration), what's stopping the og compute providers (hetzner/OVH, etc) from doing the same thing? They can do it on a much grander scale directly.
ig it all boils down to distribution eventually..
ai_agent云
@0xAsm0d3us
原文 ↗
模型量化会明显削弱漏洞利用能力,选版本与部署形态时有参考价值。
Opus 4.6 seems to have been heavily quantized at some point, but the early versions were insanely powerful.
I remember specifically seeing it just about one-shot an SPTM bypass. I looked at it and thought "there is no way that this is going to work"... And then it did
llm绕过ai_agent
@C2IRIS
原文 ↗
游戏服资源常被植入后门,CLI/CI 集成可低成本自查。
fivem-backdoor-scanner — Find backdoors in FiveM resources before you start them: hidden loaders, remote load(), obfuscated payloads, leaked tokens. CLI, CI step, library. https://t.co/kez4TPu4FK https://t.co/tuOSWpRXoY
RT @oneandonlyhusam: Day Zero, Not Zero-Day. https://t.co/MAKtDJR0HP
RT @clubby789: My DEFCON talk from earlier this year was finally uploaded today!
https://t.co/SGOfL88mAY
RT @cor_ctf: Come watch our talk at DEFCON Mainstage from this summer by @clubby789 @ryaagard and https://t.co/0NktUKF46V! Back in the days of artisanal hand-crafted Chrome exploits😇
https://t.co/s3JrC37CQE
RT @DIYprojects55: https://t.co/jeDreyUxCV
HyprBSD: A Preconfigured Hyprland Desktop for FreeBSD, Now with Steam.
HyprBSD, a new distribution that surfaced in early November 2026, aims to skip that work. It installs #FreeBSD with a preconfigured Hyprland desktop from a single ISO, and one of ..
覆盖完整引导阶段持久化手法,并扩展到新架构
While adding ARM64 support, I made a few different design choices that I will adapt to x64. Other than that, the ARM implementation follows the same principles as the x64 one: replace GRUB -> hook its Linux boot path -> hook zboot decompression -> patch a late initcall 2/3
缩小依赖面即缩小供应链与维护风险,取舍可立即执行
If you use a whole package for one small helper, consider the updates, maintenance and trust you take on. Removing unused functionality or replacing a tiny helper with simpler, tested code you can maintain may be worthwhile.
https://t.co/2nVqW3Mlsv
供应链依赖报告
@PentesterLab
原文 ↗
给出围绕 agent 编排而非换模型做优化的判断依据
RT @oneandonlyhusam: My analogy on why the Harness matters MORE than the model when it comes to Frontier Models for Security Research and RE. https://t.co/jL9KyiQmiH
ai_agentllm议题
@0xpwnie
原文 ↗
RT @OpenBinAI: Big news! https://t.co/C2KMeVjb2Q has been approved for AWS Open Source Credits!
7,000+ users. 100% organic growth. An incredible security research community.
Huge thanks to @AWSOpen for supporting collaborative security research!
https://t.co/Sbxod6EELI
RT @oneandonlyhusam: https://t.co/lRxqSx7BYh will STAY FREE Thanks to AWS!
@felixm_pw ohh pretty solid work. i like how you have pre-selectable payloads to choose from 👀
As promised in my SEC-T talk (which will be available on YouTube shortly), this is only the beginning. I want to consolidate the x64 implementation to match the ARM one, widen Linux distro support, and deal with EFISTUB (which was a great point raised in SEC-T!) 3/3
RT @qkaiser: you only need 2 things to root any access point in 2026:
1. a collection of proprietary drivers from Realtek, Ralink, Broadcom, and Mediatek
2. tokens
RT @Commsrisk: Tokyo Police Arrest Chinese National Using SMS Blaster to Scam Tourists - https://t.co/aHxHSt0QhU
Police arrested a Chinese national operating an SMS blaster in Tokyo. The device reportedly disrupted service for thousands of phones and sent smishing messages to tourists.
RT @LeCueilleur: Gasthof „Alter Wirt“, Alte Hauptstraße 49, Hechendorf, Upper Bavaria. The unassuming pension on the Pilsensee served the U.S. Counter Intelligence Corps (CIC) as a safe house for several years 1/6 https://t.co/erAWksJHX7
完整执行链可转成检测规则与狩猎线索,含十进制编码 IP 等规避手法。
RT @Kostastsale: We captured this attack as well, and managed to follow the execution chain quite a bit further.
It started with the same obfuscated PowerShell command using curl.exe and a decimal-encoded IP (𝟷𝟹𝟼𝟿𝟿𝟾𝟶𝟺𝟼𝟷 → 𝟾𝟷.𝟷𝟼𝟾.𝟼𝟸.𝟺𝟻). From there, we observed multiple stages of payload delivery, persistence, process injection, and credential access.
Here's what we saw:
1️⃣ PowerShell downloaded 𝚏𝚕𝚘𝚖𝚘𝚘𝚏𝟷.𝚣𝚒𝚙 from 𝚖𝚊𝚒𝚕𝚍𝚎𝚟𝚎𝚕𝚘𝚙𝚎𝚛.𝚜𝚒𝚝𝚎, extracted it under AppData\Local\ExFiles, and executed 𝚏𝚕𝚘𝚖𝚘.𝚎𝚡𝚎.
2️⃣ flomo.exe established persistence through the user Run key (𝚎𝚕𝚎𝚌𝚝𝚛𝚘𝚗.𝚊𝚙𝚙.𝚏𝚕𝚘𝚖𝚘), then extracted and launched 𝚄𝚝𝚒𝚕𝚒𝚝𝚢𝙿𝚒𝚡𝚎.𝚎𝚡𝚎.
3️⃣ UtilityPixe.exe copied itself to 𝙲:\𝙿𝚛𝚘𝚐𝚛𝚊𝚖𝙳𝚊𝚝𝚊\𝚑𝚊𝚗𝚍𝚕𝚎𝚛_𝚍𝚋𝚐_𝚍𝚎𝚟\ and staged a legacy scheduled task file (JSONGateway_debug.job).
4️⃣ It then spawned XenoMixe.exe suspended and modified its thread context using 𝚆𝚘𝚠𝟼𝟺𝚂𝚎𝚝𝚃𝚑𝚛𝚎𝚊𝚍𝙲𝚘𝚗𝚝𝚎𝚡𝚝.
5️⃣ 𝚇𝚎𝚗𝚘𝙼𝚒𝚡𝚎.𝚎𝚡𝚎 accessed Edge's Cookies, Local State, and Login Data files, registered raw keyboard input hooks, and injected into an active Edge process using SetThreadContext.
6️⃣ Browser data was staged into temporary directories and .7z archives. We also observed AD reconnaissance using net group "domain computers" /dom.
7️⃣ Network activity included 𝚐𝚙𝚝𝚌𝚖𝚍.𝚒𝚘 and outbound connections to multiple IPs on ports 2222/2223.
One thing worth noting is that we confirmed access to browser credential stores and data staging, but couldn't verify successful exfiltration from the available network telemetry.
IOCs:
𝟺𝟻[.]𝟷𝟹𝟷[.]𝟸𝟷𝟺[.]𝟸𝟹𝟿:𝟸𝟸𝟸𝟸 ➔ attempted a TCP connection from 𝚇𝚎𝚗𝚘𝙼𝚒𝚡𝚎.𝚎𝚡𝚎
𝟾𝟺[.]𝟸𝟻𝟸[.]𝟼𝟼[.]𝟷𝟽𝟼 ➔ XenoMixe.exe attempted TCP connections on ports 2222 and 2223
𝚐𝚙𝚝𝚌𝚖𝚍[.]𝚒𝚘 ➔ Resolved by the injected malware process 𝚇𝚎𝚗𝚘𝙼𝚒𝚡𝚎.𝚎𝚡𝚎 to establish bidirectional network communication on TCP ports 3333 and 3334, operating as C2.
𝚖𝚊𝚒𝚕𝚍𝚎𝚟𝚎𝚕𝚘𝚙𝚎𝚛[.]𝚜𝚒𝚝𝚎 ➔ Repeatedly resolved by a staging executable (flomo.exe) and served as the source for downloading malicious payloads during staging of additional components.
𝚑𝚝𝚝𝚙://𝟾𝟷\[.]𝟷𝟼𝟾\[.]𝟼𝟸\[.]𝟺𝟻/𝚓𝚎𝚜𝚜/𝚊𝚙𝚒?𝚝=<𝚄𝙸𝙳>&𝚝𝚛𝚊𝚌𝚔=𝟷&𝚎𝚟𝚎𝚗𝚝=𝚍𝚕_𝚘𝚔
𝟾𝟷[.]𝟷𝟼𝟾[.]𝟼𝟸[.]𝟺𝟻 ➔ The IP that the loader contacted
powershell事件凭据窃取payload
@virusbtn
原文 ↗
OT 设备裸露面与参数篡改手法可直接对照自查同类控制器。
Ransom-ISAC examined logs from an exposed refrigeration controller where intruders made 743 parameter changes across 31 devices and issued alarm-mute and defrost commands. The logs do not show whether the commands caused physical damage. https://t.co/gPsY4meO7o https://t.co/LgaHMySTw4
伪装开源与 KeyGen 的投递链完整,可当作社工诱饵模板参考。
RT @SlowMist_Team: 🚨 SlowMist TI Alert: Beware of the “Cracked TradingView Premium” Infostealer 🚨
Since mid-October, X account "cryptotrice" has repeatedly promoted a so-called “reverse-engineered cracked #TradingView Premium,” with some posts passing 300K views. The software requires downloading a password-protected archive and generating a license code via a KeyGen. 🔑
The author claims the project is open source on GitHub, but the download link in the article does not point to #GitHub. In some replies, the author also said the repo has been made private and that links are only available in-app and expire after a single use. Several users warned in the comments that it looks like malware stealing account info and crypto wallets, and one said their antivirus flagged and deleted it right after extraction. 🦠
We analyzed it and confirmed the Windows sample is an infostealer:
🔸 Collects browser credentials, crypto wallet data and Telegram data, and sends it to the C2 via WinINet
🔸 Uses Telegram and Steam to dynamically fetch and rotate C2 servers
🔸 The download site, arche-hypnose[.]com, appears to be a legitimate website compromised to distribute the malware
🔸 The macOS link currently returns 404, so no sample has been obtained
⚠️ If you have already run this software: disconnect from the network immediately, move your assets to a clean device with a new wallet and seed phrase, change passwords for your browsers, exchanges, email and Telegram, and terminate active Telegram sessions.
Cracked software, KeyGens and password-protected archives are among the most common ways infostealers are distributed. Don’t install them. 🛡️
IOC
Sample source: teapawt[.]org/cryptotrice/article/2111472920241524897
Download links:
hxxps://www[.]arche-hypnose[.]com/get/ (Windows)
hxxps://www[.]arche-hypnose[.]com/mac/ (macOS, currently 404)
C2:
hxxps://xn[.]amanlay[.]org/rn/gate
hxxps://z[.]dak69[.]org/j/gate
hxxps://xn[.]amanlay[.]org/kw/gate
恶意软件信息窃取钓鱼
@virusbtn
原文 ↗
经 SOCKS5 代理、以中文报错判真的布尔盲注手法可以迁移复用。
RT @Huntio: Three open directories on the same host, captured over five weeks. All of them pointing at one target: Tamkang University in Taiwan 🇹🇼
The operator built a custom blind SQLi framework against the university’s IP registration system at 163.13.240[.]200. Boolean oracle over SOCKS5, TRUE/FALSE detection via Chinese error strings.
The companion script enumerates information_schema, samples rows from 10 tables. The output file confirms it worked: 212 IP ranges, 170 support tickets going back to 2005, full network inventory with usernames, department codes, room numbers, phone extensions, equipment, OS types.
On a different port, the same server holds SEP14E64A.exe, a Windows RAT named to look like a Symantec installer. WSS C2 via SChannel, systeminfo beacon, file transfer with chunking and resume, process/registry/service enumeration, browser detection. Config is encoded behind a CFG! marker, not plaintext. Next to it: wisp_agent_linux, a Go binary with shell + SOCKS components supporting five transport modes including DNS.
Then the http.log on :8899. A TANet IP (163.13.121[.]218, MoE Computer Center TW) listed the directory and pulled bdgrab.exe with a 200 on Oct 11.
The server also stages credential stuffing scripts, SSH pivots via sshpass and Paramiko, MS17-010/EternalBlue targeting five Tamkang hosts, CVE-2023-46604 ActiveMQ payload builders, and FortiSIEM CVE-2025-25256 + CVE-2025-64155 PoCs.
No actor attributed.
多域名回连与反分析逻辑可提炼成针对 Python 恶意软件家族的检测点。
Seqrite analysed a newer Armorsteal build after finding three scheduled tasks on an infected Windows system. The Python kit uses 11 fallback domains, steals data, and stops its cryptominer whenever Task Manager or another monitoring tool opens. https://t.co/8UrmnUnBA7 https://t.co/oMElTBY1bh
恶意软件python挖矿
@virusbtn
原文 ↗
Remember: the codebase you're reviewing is alive, and it won't stop changing just because you're taking notes.
When documenting a potential vulnerability, don't record only the filename and line number. Code moves, code changes, files get renamed, and your reference might point to something completely different by the time someone reads your findings.
Always keep the filename, line number, commit hash, and a copy of the relevant code, along with an explanation of the issue. If you include a link, make sure it points to that specific commit, not to the branch.
Your notes should still make sense after the code has changed.
用 AI 管理服务器之后,宝塔,1panel 这些服务器管理面板很鸡肋了
ai_agent
@nodotbtree
原文 ↗
I understood you didn’t say without tools. I was clarifying what I meant. Anyway, what you said is just not aligned with reality aside the complexity of mobile attacks, I don't know enough about mobile exploitation to discuss; and have not applied open-weights to it. But for chaining and exploiting or full effect on common operating systems or software, yes, it is doable with open-weights. No, I'm not talking about WHQL drivers with some poorly structured IOCTL -- don't oversimplify.
Catastrophic misunderstanding of what users want in the Registry Editor.
windows
@0xTriboulet
原文 ↗
给出模型自主做利用链的能力上限,可评估自动化攻防风险。
To be clear, I did not say with no tools.
What I am saying is *with tools* -- e.g. with the headless API for the best commercial decompilers, and emulation/virtualization solutions (like Corellium Falcon), etc -- which in total, cost hundreds of thousands of dollars, and with a large sample size of working production grade exploits to take inspiration from, the best open source models are *still* completely incapable of autonomously designing and building production grade exploits/exploit chains against the major mobile attack surfaces. With the very close and careful guidance of an expert exploit developer with years of experience, they can be halfway useful.
They can find bugs, sure. But that is infinitely easier than the work involved with curating, exploiting for full effect (i.e. RCE, SBX Escape or LPE) and chaining.
However, the very best closed models, in their various cyber-vetted versions, are capable of doing so. This only became the case in the last 6 weeks or so.
红队视角的注入与免杀手法,配代码可直接演练。
https://t.co/Y5MD2shsh0 next one from my blog #windows #malwaredevelopment series. enjoy! ❤️
https://t.co/ySDpkCximF
#hacking #maldev #threatintel #research #purpleteam #redteam #blueteam https://t.co/P6MplT2vpl
恶意开发windows绕过
@cocomelonckz
原文 ↗
涉及 ETW-TI 遥测盲点,对免杀与检测对抗有参考价值。
some commercial frameworks have a drawback when using CET, but when it comes to vanguard, you don't have to worry about spoofed frames during bof executions, or AMSI and ETW evasions. although, i do take extra care about ETwTi's that feed telemetry as a snack pack haha.. https://t.co/MnIWRM5VIJ
Have a burning security question? Just @ Hacktron AI, now in Slack.
Check whether a new CVE affects your repos, ask how your code handles something, or save a decision straight to your threat model, all from the thread. It also triages findings, opens fix PRs, reviews pull requests and sets up pentests.
Get answers from your own code and threat model within minutes.
Try it yourself: https://t.co/5D3nTUEPt4
ai_agent工具
@HacktronAI
原文 ↗
RT @InstLatX64: The x86 Ecosystem Advisory Group released the #AVX10_3 and #ACE v2 (AI Compute Extensions) specifications:
https://t.co/ndYCjAHXXZ
#EAG #Intel #AMD #Zen7 #CoralRapids #AVX512 #AVX10 #AMX https://t.co/HhDRE8eh5Z
@MalwareTechBlog ofcourse, there are insane amount of low complexity bugs and misconfigs that glm and deepseek flash would exploit cheaper than frontier models.
@MalwareTechBlog i also think the reason for less exploitation of php2shell is the gap?
勒索生态内部互撕与伪造旗号,可作 CTI 溯源与虚假归因的观察样本。
RT @CTI__Updates: Someone made an onion with info doxing Qilin operators that looks like its from The Gentlemen ransomware group but then thegroup has claimed they have nothing to do with it
???
what is going on lol WE GOT RANSOMWARE GROUP DRAMA 🍿🍿🍿
hxxp://qilinjzaot34gv3ugxaqhiwomeuwqfkf7nblpyp4dr7igpn22ib6kkid[.]onion
ransomwarectidoxing事件
@0xocdsec
原文 ↗
跨境暴力犯罪与枪支走私手法的传闻,尚未经证实,仅供风险感知。
RT @izhouhongan: 卧槽!卧槽!杭州一个老板骗了缅甸人的玉料,没给钱,被跨国追杀!
缅甸人来到杭州连开3枪杀死了!
枪是拆成散件运来组装的!