⚡Megatron LLM Hub
全部
Twitter

推特安全流

头条7

签名均有效、未被任何 AV 拦截或吊销,BYOVD 可直接致盲终端防护

RT @SEKTOR7net: Reversed and Weaponized Microsoft-Signed Driver Used to Kill CrowdStrike EDR. 15+ variants of a kernel driver (PoisonX) used in BYOVD attacks against EDRs - all carrying valid Microsoft signatures, none blocked, revoked, or detected by any AV vendor. Reverse engineering revealed a single IOCTL (0x22E010) that takes a PID as an ASCII string and calls ZwOpenProcess() + ZwTerminateProcess() against PPL-protected processes. A post by Jehad Abudagga Source: https://t.co/Jr9KT6lGW9 #redteam #blueteam #byovd #edr #windowssecurity

byovdedr驱动poc @artem_i_baranov 原文 ↗

绕开 SYSTEM 服务与 COM 接口,普通用户态即可导出浏览器凭据

A reliable, simple Rust port of ElevationKatz that demonstrates how Chrome's and Edge's App-Bound Encryption (ABE) can be defeated without ever touching the SYSTEM elevation service, COM interfaces, or any elevated privilege by @L33tH4xcyber https://t.co/0TklogAfra

浏览器凭据窃取工具poc @ipurple 原文 ↗

命中特征与分支名已给出,CI与OIDC凭据泄露等同全环境失守。

RT @AikidoSecurity: 🚨 Massive GitHub Actions CI Secret Stealer 4,800+ compromised accounts attempted to push malicious workflows to 61k+ repos. Payload dumps all CI secrets, memory, OIDC & commit secrets, then instantly deletes run logs. Search your repos for: • PR: "docs: dependency audit notes" • Branch: wfx-docs or chore/ • Commits: "ci: refresh build cache" or "chore: add ci helper" • Actions tab: Skipped/missing run numbers If hit, assume total compromise. Rotate ALL secrets and cloud roles immediately.

供应链云ci事件 @virusbtn 原文 ↗

本地逆向 Agent 的实用样板:先审批后执行,结论需带证据

IDA Assist 1.1.0 is out. Give a binary a goal and the agent plans the work, waits for approval, then backs every claim with evidence from the database, across sessions and restarts. Also: approval dialogs for MCP tool calls, query and control-flow cards, multi-anchor Memory.

ai_agent逆向工具mcp @in19h 原文 ↗
必看17

自建 ScreenConnect 服务器在受影响范围,蠕虫可横向扩散至下游已连客户端。

Today (1) we saw another wave of SCKit compromises against ScreenConnect servers and downstream connected clients, including on-premise instances via the weak RMM+ plugin. Also, (2) I chased "cryptobait" fake guest connections that appeared in multiple ScreenConnect servers. 🔗👇 1. https://t.co/zQOu3Rl0av 2. https://t.co/a6qFikeueK

rce供应链事件 @_JohnHammond 原文 ↗

植入件与安全元件同板,外观与功耗异常都难以察觉,供应链检测思路需更新。

Hardware implants in retail crypto wallets are usually spotted by visual inspection or anomalous power draw. The Trezor Safe 5 units sold by one former reseller contained an added radio that exfiltrated the seed over cellular. The implant sat on the same board as the secure element. Our CryptoBilis investigation: what we found and what we’ve changed | Trezor https://t.co/1DPoJ7POCw

供应链固件事件 @0xpwnie 原文 ↗

滥用合法签名驱动的 BYOVD 绕过,驱动黑名单与内核监控可直接据此加固。

RT @_josehelps: The Gentlemen ransomware crew keeps bringing its own drivers. @Acronis TRU found their EDR killer loading Safetica and Lenovo BootRepair drivers to kill security tools from the kernel. Great work by @spontiroli and team: https://t.co/Tfc553A1rY

byovd内核事件 @0xTriboulet 原文 ↗

LLM 驱动的智能体已被实战用于串联攻击链,自托管同类平台的暴露面需重新评估。

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 @Teach2Breach 原文 ↗

头部研究团队的披露窗口变化会牵动厂商补丁节奏与防御方的自查计划。

RT @natashenka: In the age of AI, 90 days isn’t fast enough. Today, Project Zero updated our disclosure policy to reflect the changes needed to protect users in this new world. https://t.co/7xGtVypUga

漏洞披露报告ai @moyix 原文 ↗

凭据窃取比漏洞利用更常成为突破口,检测重心与演练假设需相应调整。

RT @snehalantani: Attackers don't "hack in" with CVE's or zero days, they log in with credentials they've compromised. As security practitioners this isn't news, however given the vulnpocolypse craze and maniacal focus on squashing application bugs, it's a principle that's important for all to remember.... and here's a clear example from a real customer environment... our AI Hacker NodeZero was run against a large enterprise. No humans in or on the loop, no prior knowledge of the environment, just point+click+hack During that pentest (or, "agentic attacker swarm" is what the cool kids are calling it now 🤣 ), NodeZero methodically compromised credentials using a variety of techniques. In this case, NodeZero: - successfully cracked 61k+ credentials due to weak passwords - and it took ~5 mins of GPU time to crack a single credential - A single credential provided access to 11k hosts (!!) due to credential reuse - 1300+ credentials were compromised because the EDR on the host wasn't configured properly (happens all the time) - 1200+ credentials had easily guessed passwords This is the attacker's perspective... so while organizations surge resources into fixing vulnerabilities in custom code, remember the attackers have a vote... and it's much easier for attackers to reuse credential compromise techniques across thousands of targets versus spending a ton of time researching a zero day in your custom app that cannot be reused elsewhere #ai #hacking #pentesting #infosec

ai_agent凭据事件 @0xTriboulet 原文 ↗

规避型载荷托管的新思路,检测点明确,可自行复现验证。

Interesting covert technique released by @S3N4T0R_0X0 earlier this year. (No time to test earlier!) Implants are stored within Pixel Data (.MP4). I did some minor modifications to the original PoC and used Havoc as the test case. Platforms such as YouTube could potentially be used as remote storage. Planning to test the detection opportunities this weekend.

隐写逃逸poc免杀 @ipurple 原文 ↗

后门在Agent中可触发工具调用与命令执行,且不依赖精确触发串。

RT @MsftSecIntel: A model backdoor embeds hidden behavior that stays dormant until a trigger appears. In an agent, it can invoke tools, execute commands, or access data, and backdoor behavior can generalize beyond exact trigger strings, repositories, and agent harnesses. https://t.co/DNg23W8t0Z Two studies of the post-training supply chain, from Microsoft Defender Research and the Microsoft AI Red Team, examine poisoned fine-tuning datasets and backdoored open-weight checkpoints. Their findings support Satya Nadella's argument that trustworthy enterprise AI requires independent observability and controls that sit outside the model. Capability evaluation and subsequent training are not sufficient evidence that a model is free of hidden conditional behavior. Read the blog for the full findings and guidance on controls over training-data pipelines, model adoption, and tools.

ai_agentllm供应链隐藏后门 @virusbtn 原文 ↗

从CI密钥横向到云IAM持久化,完整链路与还原细节已公开。

Aikido found a GitHub campaign stealing CI secrets and using them to seek cloud access. The wfx payload could rewrite AWS trust policies and add Azure federated credentials, while attackers reverted commits and deleted workflow runs. https://t.co/l5i1yS84Yd https://t.co/jd3Ck5cy8w

供应链云iam报告 @virusbtn 原文 ↗

开启失窃设备保护仍可解锁,疑似已静默修补的蓝牙漏洞

RT @IntCyberDigest: ‼️ A leaked Cellebrite flyer shows forensic labs a remote iPhone unlock appliance with a range of about 8 metres. It covers the iPhone 11 to 16, even with Stolen Device Protection on. The blogger who got it suspects a Bluetooth bug Apple quietly patched. https://t.co/HwEjScUqMP

移动端取证蓝牙事件 @0xocdsec 原文 ↗

默认配置的 Defender 即受影响,邮件、网络盘、U 盘皆可触达,PoC 已公开。

RT @killa: what if "Windows Defender" was actually "Windows Offender"? introducing CupidsArrow, sometimes falling in love with a file and inspecting it on first sight isn't a good idea... it is not filetype specific. it could be a jpg, mp3, any type of file and location doesn't matter either, could be from an email, networkdrive, usb drive... i wanted to add some fun stuff but im too tired and sick to do it... also @JensenHuang is a legend (sorry for the meme in the poc) and i agree with the nvidia employees, bring back THE boss! happy friday, have a great weekend and goodnight (and yes, microsoft will sort it out once they fix their already broken services and my video editor was sleeping so no fun editing stuff) xoxo

windowsdefenderpoc免杀 @weezerOSINT 原文 ↗

自托管 Agent 直接握有账号级令牌,攻击面从模型转移到插件与凭据。

RT @matthewsoldit: introducing log our tool to let your ai agent to become you. so you can do the things you actually enjoy. log powers gives your harnesses the power to find opportunities in every conversation you've ever had. we built out our intergration into @NousResearch hermes agent as a plugin and it's all opensourced! log is built on top of @treg_ai @subquadratic and @morluto's reverse engineer anything to find the internal api's of your social medias (or any website) so your agent can just take control.

ai_agent工具逆向凭据 @morluto 原文 ↗

一条查询即可在 Defender XDR 中筛出可被远程利用的 Apache 主机。

🚨 Hunting for internet-facing Apache servers vulnerable to CVE-2026-119618 (CVSS 9.8)? ​ This KQL combines 𝙳𝚎𝚟𝚒𝚌𝚎𝙽𝚎𝚝𝚠𝚘𝚛𝚔𝙴𝚟𝚎𝚗𝚝𝚜 and 𝙳𝚎𝚟𝚒𝚌𝚎𝚃𝚟𝚖𝚂𝚘𝚏𝚝𝚠𝚊𝚛𝚎𝚅𝚞𝚕𝚗𝚎𝚛𝚊𝚋𝚒𝚕𝚒𝚝𝚒𝚎𝚜 to quickly identify exposed Apache instances running vulnerable versions across both Linux and Windows. 𝗜𝗱𝗲𝗻𝘁𝗶𝗳𝘆. 𝗜𝘀𝗼𝗹𝗮𝘁𝗲. 𝗣𝗮𝘁𝗰𝗵. ​ KQL Code: https://t.co/DLkdWKW06E ​ #DefenderXDR #KQL #ThreatHunting #Apache #DetectionEngineering

cveapache检测rce @0x534c 原文 ↗
推荐15

默认安全项变更会打断依赖未签名 NTLM 的旧应用与扫描工具。

RT @abdo_mhanni: The new NTLM signing security improvements in Windows look great and are long over due. While all supported Windows Server versions are capable of NTLM signing; The new changes/update help ensure that signing is negotiated and eventually required by default. Phase 1 (November 10, 2026): Windows client devices begin offering NTLM signing on all connections by default. Servers accept both signed and unsigned clients. Auditing events identify clients not using signing. Phase 2 (May 2027): Windows servers begin requiring NTLM signing by default. Clients that do not offer signing will be rejected by secure servers. Nice to see Microsoft begin making changes that don't just address single vulnerabilities but address entire classes of attack paths https://t.co/hYoDVBhcXb

认证windows加固 @0x64616e 原文 ↗

同期给出模拟与 4104、Sysmon 检测项,可直接落地到检测工程。

RT @ipurple: 🎙️ I wrote an article today to approach ScriptBlock Smuggling that is used for AMSI evasion. The article is balanced between emulation and detection guidance for cyber defense teams. 𝑫𝒆𝒕𝒆𝒄𝒕𝒊𝒐𝒏 𝑺𝒕𝒓𝒂𝒕𝒆𝒈𝒊𝒆𝒔 ✅️ Module & ScriptBlock Logging | 4103 & 4104 ✅️ Image Load | 7 (Sysmon) ✅️ File Creation | 11 (Sysmon) ✅️ Portable Executable | 29 (Sysmon) ✒️ https://t.co/KucVmlQqcy

绕过powershell检测 @ipurple 原文 ↗

仅需贡献者权限即可篡改页眉页脚,WordPress 站点面覆盖广。

RT @LazyTitan33: I got a new CVE-2026-127094 High (7.6) in WordPress plugin called Pagelayer. Missing Authorization in Abilities API in >= 2.1.7, < 2.2.1 Contributor-level attackers can CRUD all pages+posts, & modify site-wide theme templates like headers and footers. https://t.co/9Pf4M1103n

cvewordpress授权绕过 @0xocdsec 原文 ↗

固件层口令防护可被离线绕过,大量二手与旧设备同受影响

Turns out a forgotten BIOS password does not necessarily lock you out forever 💀 We locked ourselves out of one of our older notebooks, a T450s. There are several services online that will remove the password for a fee, all of which seemed a bit sketchy 🫠 https://t.co/l1lxvEwuDV

固件绕过报告 @RedTeamPT 原文 ↗

集成 PE 解析、Capstone、capa 与沙箱引爆,可直接上手试用

Vyrulix — Open-source, bare-metal desktop malware triage and static threat intelligence workbench for Windows (MapViewOfFile PE parsing, Capstone disassembly, capa MITRE mapping, and 15s Windows Sandbox detonation). https://t.co/7QRjF18qE1 https://t.co/soinxOLQna

工具恶意软件静态分析 @KitPloit 原文 ↗

用区块链做死信箱绕过静态检测,目前多数引擎静态免杀。

RT @moonlock_lab: 1/ We analyzed few stages of a Node.js #macOS loader that hides its next stage on public blockchains and pulls it only at runtime: Stage 1 → TRON memo → BSC tx → Stage 2 → HTTP C2 → Stage 3 FUD on VT static engines as of now, shared by @malwrhunterteam. More info below👇 https://t.co/nIBWnYcRI7

macos免杀恶意软件区块链 @virusbtn 原文 ↗

无需入站监听、共享密钥替代C2地址,PHP与cPanel环境命中最多。

RT @Huntio: ⚠️ Hackers Abuse GSocket to Maintain Persistent Access to Linux Servers https://t.co/JkOokWTU32 Hackers are deploying GSocket on compromised Linux servers to maintain access even after the initial vulnerabilities are patched. GSocket uses encrypted outbound connections through a relay network, with a shared secret replacing traditional C2 addresses. No inbound listener is needed. Most detected activity involved PHP applications and cPanel hosting environments, with payloads staged under /dev/shm and disguised as system processes. Persistence involves cron jobs, systemd services, and shell startup files. Some attacks also attempted to exploit CVE-2026-43284 for root access. #ThreatIntel #ThreatHunting #GSocket #Linux #CyberSecurity

linux后门c2事件 @virusbtn 原文 ↗

边界设备失守后直接触及虚拟化层并造成破坏性清除,链条完整。

Check Point Research investigated an intrusion that began with PHP2Shell exploitation on a security gateway. The attackers installed backdoors, scanned the internal network and moved over SSH to an ESXi host, where they ran a wiper. https://t.co/C0AxzeOBTQ https://t.co/1F8z124zwB

rce横向移动事件报告 @virusbtn 原文 ↗

自建安全 Agent 选型参考:检索增强胜过微调,开源模型落后

these are the systems I tested: - vanilla open-weight model, no fine-tuning - open-weight model, fine-tuned on security corpus - open-weight model + (RAG)truth oracle for security corpus - Frontier model, no oracle - Frontier model + oracle guess which one performed better? across multiple runs. 🥇Frontier model + oracle 🥈Frontier model, no oracle 🥉open-weight model + (RAG)truth oracle for security corpus yes, open-weight model, fine-tuned on security corpus came at last place. So if you must use open-weights models, you are better off designing a truth oracle for security corpus with some v good retrieval/ranking system, instead of fine-tuning.

llmai_agent报告 @0xAsm0d3us 原文 ↗

覆盖 Windows/macOS/Linux 与 iPhone 备份,双击即用,支持 27 种语言。

Zero-Breach-Toolkit v2.0.7 — Free, open-source check for traces of mercenary spyware (Pegasus, Predator, Graphite, Candiru and more) on computers, Android phones and iPhone backups. Double-click apps for Windows, Mac and Linux, in 27 languages. https://t.co/6G5DqGdw9K https://t.co/0noG3AprhI

工具移动端取证 @KitPloit 原文 ↗
更多70

宏、ISO、LNK 被封后的现用执行链,2018 年的漏洞至今未修,手法可复现。

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.

xss绕过事件 @0xpwnie 原文 ↗

隐藏 opcode 是固件与虚拟化逃逸的盲区,工具已公开,可本地跑。

RT @0xpwnie: CPU designers intentionally hide instructions from the public documentation. They are called "non-standard instructions" and they exist in every modern CPU. A 2025 fuzzing tool called MystFuzz systematically hunts for these hidden opcodes to find security bugs before attackers do.

工具固件poc @0xpwnie 原文 ↗

AI 自动找洞与给复现步骤已产品化,可作为自建 agent 能力对照基线。

RT @nickvangilder: Having used both XBOW and Armadin, Daybreak Red Team seems like a pretty obvious competitive play by OpenAI to capture some of that lucrative AI-powered penetration testing market share. At their core, all 3 products do many of the same key things (find vulns, demonstrate exploitability at runtime, provide steps to reproduce, offer remediation guidance, etc.) but Daybreak Red Team looks like it’s trying to take it even further by integrating with Daybreak Defense Factory and essentially offering a closed loop solution that can also write a code patch via Codex, open a GitHub PR, and retest the fix. I predict XBOW and Armadin will be announcing similar capabilities soon.

ai_agent工具事件 @0xocdsec 原文 ↗

源码可审、可本地部署的隐私网络工具,适合自建隐蔽通道评估。

I2Pplus — Java soft-fork of the I2P anonymizing network layer, providing an onion-routing router, tunneled services, and bundled I2PSnark BitTorrent client for private networking. https://t.co/e5Dx3S1uNz https://t.co/3JCOfVrown

工具自托管 @KitPloit 原文 ↗

RT @0xpwnie: During Windows 3.1's installation setup, Microsoft embedded self-modifying XOR-encrypted code that secretly checked whether the machine was running DR-DOS instead of MS-DOS. If it detected a competitor's OS, it deliberately triggered a fake error screen to scare users away. The engineer who wrote it embedded his own initials in the encrypted signature. It was discovered years later during antitrust litigation. Microsoft paid $280M to settle.

逆向事件 @Teach2Breach 原文 ↗

RT @0xpwnie: This just reminded me of a happening i heard of back when i was in junior school: The Night Shot Flaw aka see-through camcorders Sony’s consumer Handycam camcorders (models like the CCD-TRV65 and others with the NightShot/NightShot infrared mode) could record in very low light or darkness by mechanically moving aside the normal IR-cut filter and using near-infrared sensitivity (plus built-in IR LEDs for illumination). The mode produced the familiar greenish monochrome image. Under very specific daytime conditions; bright sunlight (which contains a lot of near-IR), certain thin/dark or loosely woven fabrics (especially some swimsuits, black synthetics, or sheer materials), often with an added visible-light-blocking IR-pass filter over the lens, and the right angles/distance, the clothing could appear partially translucent in the IR image. This revealed body silhouettes or underwear outlines rather than true X-ray detail or full nudity. It worked because some fabrics absorb/reflect visible light differently from near-IR wavelengths. A Japanese men’s magazine highlighted the effect, it spread quickly, and Sony confirmed limited reproducibility under those constrained conditions. They halted further shipments of the affected models, modified the feature (so NightShot would not engage in sufficient ambient light on later units), and did not issue a full recall of the hundreds of thousands already sold. Demand for the unmodified cameras spiked briefly because of the publicity. It was never reliable “X-ray vision,” required ideal lighting/fabrics/setup, and was more of an unintended side effect of the IR sensitivity than a designed flaw. Similar IR transparency effects can still appear with certain modern night-vision or modified cameras under the right conditions, which is why some athletic wear (e.g., for the Olympics) has been developed with IR-blocking properties.

硬件隐私 @0xpwnie 原文 ↗

RT @HackingLZ: Ransomware forced all kinds of organizations to start caring about cybersecurity when they previously didn’t. Infosec is still a cost center for most businesses. People seem to be forgetting the plot here. The goal shouldn’t be more infosec. It should be secure defaults that reduce the need for it in the first place.

议题 @0xTriboulet 原文 ↗

免手工编写的堆行为分析工具,可直接用于恶意软件取证与检测验证。

Claude just made this tool, HeapExplorer, hooks Windows heap functions to show realtime memory allocations, detect encrypted buffers, extract strings, and spot implant patterns like key rotation and unbacked shellcode I can upload to GitHub if anyone is interested in the tool. https://t.co/9JsWyqLtav

工具内存取证windows @papadelospapas_ 原文 ↗

C2 隐蔽基础设施的实操教程和代码更新,红队可直接复用验证。

https://t.co/zKV2rbGXx9 updates today! - blackmagick-c2 update: new version includes enhancements for use with CDN & guide - Video tutorial for setting up c2 with CDN redirector - New Malware Pulse article: LACUNA Leaf-gap frames - New tech: BM-T2013 Leaf Gaps code & tutorial

c2工具绕过 @Teach2Breach 原文 ↗

故障窗口是凭据钓鱼的高发期,可据此安排临时监控与用户提醒。

Every major SaaS outage creates a short window where users will accept a login prompt they would normally ignore. Microsoft confirmed the November 2026 M365 incident was a network capacity failure, not an intrusion. The verification of that claim happens inside Microsoft’s incident process. Outside observers only see the status page and the subsequent phishing volume. Microsoft 365 Outage Disrupts Outlook, Exchange Online and Teams - SecurityWeek https://t.co/jhbKZpVMjG

钓鱼凭据事件 @0xpwnie 原文 ↗

AI 语境下的漏洞利用一手研究,议题细节值得回看原始材料。

RT @unpromptedconf: Our favorite person Bruce Schneier stood in the audience at @unpromptedconf listening to this talk and asked people around him: "Who is this??" because the talk was THAT GOOD. Today’s daily dose of [un]prompted was from Anna Groza, a @calif_io security researcher on this exploit. https://t.co/FRO9HcfXfV

议题漏洞利用ai @moyix 原文 ↗

RT @Microsoft: We are deeply sorry for the disruption this incident has caused our customers and partners. We know the trust they place in our services and take that responsibility extremely seriously. We are seeing broad recovery and stability across our services. This was a technical issue not related to cyber or AI / agents. The incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change. We have addressed the underlying issue and continue to restore service to remaining customers and regions. Given the scale of demand, we restored services gradually, validating stability at each stage. While this took longer, it helped ensure a stable restoration. We will provide a full post-incident report once our review is complete.

事件云 @artem_i_baranov 原文 ↗

严重性提级的具体推理链条,对漏洞报告与影响评估有参考价值。

RT @00xalr: Alhamdulillah, Another $10,000 bounty for a Critical-severity vulnerability with a CVSS score of 10.0 🔥 Here's a detailed write-up on how I escalated the severity of a vulnerability. https://t.co/rr7AZBWu29 #cybersecurity #bugbounty #bug #hackerone https://t.co/1551gj2zlf

bugbounty报告漏洞 @0xocdsec 原文 ↗

AI 自动挖洞赛道获前沿实验室背书,可据此判断工具链演进方向。

RT @oegerikus: When we started @Xbow in January 2024, believing that AI could autonomously exploit vulnerabilities like a skilled security researcher. People laughed in our faces. Since then, we have been working closely with the teams at frontier labs to realize our vision and find bugs in software before the bad guys do. Yesterday, @OpenAI reaffirmed our vision, positioning offensive cyber security as the most consequential application of AI alongside coding. Our market has gone mainstream. The problem of security is now so big that we must tackle it with great urgency across the industry. We need all defenders on deck, now! XBOW is built for the enterprise, where integrations, controls, and safety matter most. In an ideal world, customers would run aggressive offensive penetration testing against their staging environments that mirror production. The reality, however, is that these environments are often either non-existent in large enterprises or fail to accurately reflect the complexity and conditions of production. Building a system that can test production safely is the hard problem we’ve spent nearly three years solving. XBOW is model-neutral. We use the best model for each task and combine their strengths where it helps. That lets us keep improving the product as the labs keep improving the models, and it’s why we can serve the world’s most demanding enterprises today. This is all possible because of the team we assembled. Our offensive security researchers are simply the best. Just ask our customers! XBOW was founded on the idea that defenders must always be one step ahead of attackers. We can do it, and therefore we must. If attackers have access to the same models and capabilities, it’s imperative that every organization be able to act swiftly with the power of offense in their hands. We were delighted to be mentioned in the keynote, and look forward to pushing forward the frontier of cybersecurity, alongside our partners at @OpenAI , @AnthropicAI , @SpaceXSI and elsewhere. There is a great deal still to do. But moments like this one make me confident that we as an industry are headed in the right direction.

ai_agentllm事件 @moyix 原文 ↗

梳理苹果各级缓解机制被逐层突破的脉络,移动端攻防的参考坐标

RT @POC_Crew: 📱 Let’s talk jailbreaking! Lars Fröder (@opa334dev) - “iOS Jailbreaking: From P(PL)ast to PreS(PTM)ent” #POC2026 https://t.co/gGyG6cz3Ex

移动端越狱议题 @0xocdsec 原文 ↗

给出样本关联与 C2 复用线索,PID 伪装手法可直接借鉴检测

Analysis of the APT backdoor. Includes pivots to other copies and C2: https://t.co/HuOZSEIU46 Nice use of PID spoofing and other classic techniques. Policy lure targeting SE Asia Same campaign over several months. https://t.co/I58ZRnkGMT

apt后门钓鱼报告 @SquiblydooBlog 原文 ↗

@BAO0624 @skocherhan @smica83 @malwrhunterteam @tdatwja @JaffaCakes118 @JAMESWT_WT @James_inthe_box @byrne_emmy12099 This blogpost analyzes the sample and points out related ones and other C2: https://t.co/HuOZSEIU46 Though the post is AI generated, I used my expertise to guide it, correct it, tweak it to be clear, and validated all it's work. https://t.co/GZM55ySLdc

ai_agent报告 @SquiblydooBlog 原文 ↗

Recently been speaking to people in the security industry across different orgs, and one thing I'm always reminded of is how much hackers love to share personal stories of people they have worked with/for. The tales of epic work, but also funny anecdotes from conferences when you shared a coffee, or training where they sat across from you. The stuff that doesn't make the Twitter highlight reel. It is awesome to hear when they have good things to say about someone who you have only interacted with over socials. It makes me smile every time to find out that someone is just as kind and epic without a computer too. But one bit of advice for anyone who is getting started in this industry, your public persona will often get you an interview, but people will share tales about you behind the scenes if you are a dick. Treat people well.

行业文化 @_xpn_ 原文 ↗

同批次样本与C2清单已列出,可直接扩成IOC用于本地检测。

@BAO0624 @skocherhan @smica83 @malwrhunterteam @tdatwja @JaffaCakes118 @JAMESWT_WT @James_inthe_box @byrne_emmy12099 Thanks for the tag! This is interesting. I'll share an analysis, but definitely found other related samples and C2 too. - IPRD-2026_Invitation_Letter - MOU USI-INDSR, TAIWAN - Nepal-India_Hydropower_Projects - MoU_USI_India_and_INDSR_Taiwan

钓鱼ioc事件 @SquiblydooBlog 原文 ↗

移动端木马内置模型做数据筛选,检测与取证窗口被进一步压缩。

RT @Zimperium: Our latest #zLabs research reveals how ArcticRAT uses AI to analyze stolen financial data, identify banking apps, and extract sensitive information, introducing new risks to mobile security. Read the full zLabs analysis: https://t.co/lPEDNCoQI9 #MobileSecurity #AndroidMalware https://t.co/W2juzkCBlW

移动端恶意软件ai报告 @virusbtn 原文 ↗

已备好677人的定向诱饵,仅一场视频会议即可切入高价值目标。

RT @IntCyberDigest: ‼️ North Korean hackers are cloning the faces and voices of founders and CEOs from their YouTube interviews and conference talks, and planting them in fake Zoom calls. They build the fakes with mainstream AI tools such as Grok Imagine and GPT-4o. US cybersecurity firm Arctic Wolf can see the hackers' own stash of fake videos and lures. It found 40 deepfakes there and 23 more it suspects. The same files show the group, known as BlueNoroff, prepared lures against 677 people in 58 countries, 84% of them founders and executives. Each person they hack is secretly filmed on webcam, and the footage feeds the next fake meeting.

deepfake社工事件 @virusbtn 原文 ↗

在合法签名进程内执行、走官方回传通道,终端侧极难察觉。

Bitdefender Labs found a LuaJIT stealer disguised as a log file and delivered through NIKKE’s server-pushed task channel. It ran inside the legitimate launcher and used its telemetry connection to send stolen data. https://t.co/aM36blkvRO https://t.co/2VL1AStDuj

恶意软件lua供应链事件 @virusbtn 原文 ↗

A guy asked why use Heap Encryption if he don't manually allocate memory on the heap Well, STL objects like: std::string c2_domain = "192.168.1.100:4444"; May allocate memory on the heap The same applies to STL containers (std::vector, std::map), new, malloc and compiler metadata https://t.co/jlAQ8Ub7jU

免杀windows恶意开发 @papadelospapas_ 原文 ↗

GPU 只读挂载常被当作隔离边界,构成新的逃逸面

RT @POC_Crew: 🎮 Let’s talk GPU security! Pan Zhenpeng (@Peterpan980927) & Bingjhong Jheng (@st424204) - “Read-Only Is Just a Suggestion: Escaping Containers Through the NVIDIA GPU” #POC2026 https://t.co/pDVYPRGHqI

逃逸容器gpu议题 @FuzzySec 原文 ↗

常见正则写法的实测绕过样例,可直接自查同类校验

Why use ../../ when ../ is enough... The check looks reasonable: /../ means someone is walking up the path, let's return a 404! But do you really need ../../: request $page verdict --------------------------------------------------------- /static/../../../𝗲𝘁𝗰/𝗽𝗮𝘀𝘀𝘄𝗱   ../../../𝗲𝘁𝗰/𝗽𝗮𝘀𝘀𝘄𝗱 BLOCKED /static/../munin.conf ../munin.conf PASSES The first / gets consumed by the first regular expression, so $page is ../munin.conf and doesn't contain /../ but it still allows you to get what matters... Sometimes, you don't need a full directory traversal... just one directory up is all it takes.

绕过web议题 @PentesterLab 原文 ↗

身份与生物特征整包出售,属高敏感度泄露样本

[DATA LEAK] "Suc1948237" claims to sell Dyson employee and internal data on PwnForums. • 🇬🇧 United Kingdom • Dyson, a consumer electronics manufacturer. • Two packages advertised: a full employee dataset and a complaints/violations dataset. • Employee package allegedly includes passports, IDs, biometrics, and bank/credential dumps. • Priced at $20,100 and $18,100; many units marked sold out, with restocking claimed. ✔ Posted on PwnForums by "Suc1948237", the actor offers a proof package and attached sample, accepts XMR and BTC with escrow, and restricts communication to PGP only. The complaints dataset is organized by country, with the UK among those listed. 📌 S2W note: The bundling of ID documents, biometrics, bank credentials, and contact channels points to resale intent aimed at fraud, deepfake, and phishing operations. Follow @S2W_DailyThreat #DataLeak #Dyson #UnitedKingdom #DarkWeb #S2W

事件数据泄露凭据 @S2W_DailyThreat 原文 ↗

商业监控工具采购方名单获印证,可对照核查暴露面

RT @citizenlab: A new @Bloomberg investigation into Penlink’s ad-based surveillance system, Webloc, finds evidence that confirms Citizen Lab’s reports of potential Webloc customers. Senior researcher @WolfieChristl spoke to Bloomberg about the mass surveillance tool being used by security agencies worldwide. Read: https://t.co/nk4a8c6H3w

监控报告隐私 @NetAskari 原文 ↗

含可复用 IoC 与样本命名线索,便于回溯自有日志

RT @malwrhunterteam: On Tuesday, Check Point Research published an article, titled "Not All That Glitters Is Gold: Tracking CopperDragon’s Sparrow Framework Around the Globe". Too many IoCs to go through all, but thankfully there is an easy one. They mentioned a "Strategi_AS_Referensi_April2026.rar" named sample without the hash that was not even needed anyway, as that name and the details were enough to confirm that I already gave that to @smica83 for uploading in April: https://t.co/defkSIj5Q4 🤷‍♂️

aptioc报告 @cyb3rops 原文 ↗

覆盖 BOF 从编译到内存执行的关键环节,组件即将开源

RT @NtRajkit: 🐝 🪼 Inside BOFs, Crystal Palace, and JellyBee PIC (1/3): How the Loader Prepares and Executes a COFF Object in Memory What happens between compiling a BOF and executing it in memory? The compiler produces a COFF object. The implant’s loader still has work to do. 🎥 Watch the first episode: https://t.co/NQju8cbvgB The release of the first JellyBee components is approaching. One will be released under GPLv3, while the others will be available free of charge under a license provided upon request. Before the first release, this technical series explores the architectural differences behind how implant capabilities are built, loaded, and executed. JellyBee is the only command-and-control ecosystem built around a modular, position-independent implant architecture with an integrated environment for building its capabilities directly from C projects and repositories. Swarm Factory makes this architecture practical for operators by handling PIC complexity through an advanced compilation driver integrated into its build environment. Capabilities are built as independent modules within the same ecosystem that manages their execution, allowing JellyBee to maintain a small implant core. Craft & Control. To explain this architecture and how it differs from other approaches, the series examines three models: ① BOF / COFF Compilation produces the object, but loading still requires work such as applying relocations and resolving symbols and dependencies. In the Beacon BOF model, the implant provides the loading infrastructure. Modules can also use Beacon’s internal APIs and Dynamic Function Resolution (DFR) for Windows API imports. The first video follows this process in memory. ② Crystal Palace Linker The second episode explores linking and composing capabilities for position-independent execution, examining where loading and resolution responsibilities reside in the resulting design. ③ JellyBee’s Modular PIC Architecture The third episode presents the build workflow integrated into Swarm Factory, from C projects to position-independent modules. Instead of continually expanding a single implant binary, JellyBee keeps capabilities in separate modules. Its base implant currently ranges from approximately 23 to 45 KB and does not require a separate stager. Code originally developed as a BOF can also be adapted to conventional C and built through Swarm Factory into a position-independent module. Start with Chapter 1 to see what the BOF loader does in memory and understand the foundation for the next two episodes. 🧪 https://t.co/SdVLGEzzgg #Cybersecurity #RedTeam #BOF #Offensive #PIC

工具bof免杀逆向 @ipurple 原文 ↗

补充 TraderTraitor 行动脉络与完整事件分析报告

Always great to hear old friends sharing their insights on a podcast. Even better when they mention my name! Really appreciate it.🙏 Let me add my two cents on `TraderTraitor` from my personal perspective. @craiu, you might remember `AppleJeus`, one of the earliest Lazarus campaigns targeting macOS users that we discovered back in 2018. Around that time, we started seeing Lazarus-associated activities emerge as several distinct clusters. `AppleJeus` was one of the campaigns we investigated, and the attackers continued using similar tactics, trojanized cryptocurrency trading applications, over the following years. In 2022, CISA introduced the `TraderTraitor` name for a series of malicious cryptocurrency applications that shared similarities with earlier AppleJeus operations. Since then, `TraderTraitor` has become a widely used name, while different security vendors have tracked overlapping activities under names such as Jade Sleet, UNC4899, Slow Pisces, and Pressure Chollima. Honestly, I lost visibility into some of this group's activities for a while. However, they never stopped targeting the cryptocurrency industry. Their operations evolved to include fake recruitment, malicious take-home coding assessments, and trojanized developer tools involving npm, PyPI, Ruby, and other ecosystems, contributing to several major cryptocurrency thefts. One case that particularly caught my attention was the KelpDAO incident, documented in LayerZero's investigation report: https://t.co/T7IbKEK26L The malware infection chain itself was interesting, but the post-exploitation activities were what really stood out to me. The attackers demonstrated a clear understanding of cloud infrastructure, internal workflows, and how cryptocurrency businesses operate behind the scenes. Unlike the more opportunistic, fire-and-forget campaigns we've observed from some other DPRK actors, `TraderTraitor` seems to favor a big-game hunting approach, carefully selecting high-value targets and investing considerable effort in post-compromise operations. For me, that's what makes this group particularly interesting. Even after years of tracking DPRK threat actors, I still find something surprising whenever I take a closer look at this group's operations.

aptmacos报告 @unpacker 原文 ↗

Jessica Payne was a close friend of mine for years when I lived in WA. We often coordinated trips to customers just so we could bitch about terrible security practices, share war stories, share blue team vs red team tactics, or admire threat actors. I'm still in shock over her passing. She was one of the most brilliant minds this community had and we all should take the time to go through the vast knowledge that she shared to make sure she stays with us. She was also a huge fan of random facts, so in her memory please find someone to share a random fact with daily, like we did for several years.

社区 @Laughing_Mantis 原文 ↗

[DATA LEAK] "samleego" claims to sell a 1.4M-record PhonePe customer database on DarkForums. • 🇮🇳 India • PhonePe, a digital payments provider. • ~1.4M customer records allegedly offered for sale. • Includes names, linked bank, email, phone, and address. ✔ Posted on DarkForums by "samleego", with a Telegram channel and contact handle attached for orders. The listing embeds a sample of records pairing customer PII with linked bank names. 📌 S2W note: The combination of banking linkage, contact numbers, and full addresses points to resale for fraud or phishing targeting. Follow @S2W_DailyThreat #DataLeak #PhonePe #India #DarkWeb #S2W

数据泄露事件 @S2W_DailyThreat 原文 ↗

[DATA LEAK] "Naomi" claims to sell BSI Bank employee data on DarkForums. • 🇮🇩 Indonesia — BSI Bank (Bank Syariah Indonesia), a sharia bank. • Listing offers internal employee records. • Fields include names, emails, office/home/mobile numbers, home addresses, job titles, departments, and employee IDs. • A data sample is provided via a Telegram contact. ✔ Posted for sale on DarkForums by "Naomi", with buy-now and sample access routed through Telegram channels. The attached sample lists employee directory entries spanning multiple branches and regions across Indonesia. 📌 S2W note: The combination of staff names, corporate emails, job titles, and direct phone numbers lends itself to targeted phishing and BEC against bank personnel. Follow @S2W_DailyThreat #DataLeak #BSIBank #Indonesia #Finance #S2W

数据泄露事件 @S2W_DailyThreat 原文 ↗

[INITIAL ACCESS] "BurgerDiplomatSki" claims to sell Thai Airways Pakistan email access on DarkForums. • 🇵🇰 Pakistan • Thai Airways Pakistan, the airline's local operations. • Live email account access allegedly offered for sale. • Mailbox exceeds 17GB, handling active inbound and outbound mail. • A proof-of-access screenshot is offered on request; price by direct message. ✔ Posted on DarkForums by "BurgerDiplomatSki", who frames the access as high-quality with customer and company data. The actor routes buyers to multiple encrypted channels and withholds pricing and proof until private contact. 📌 S2W note: Active mailbox access of this size could enable business email compromise and downstream fraud against the airline and its contacts. Follow @S2W_DailyThreat #DarkWeb #Pakistan #Thaiairways #S2W

初始访问事件 @S2W_DailyThreat 原文 ↗