⚡Megatron LLM Hub
全部
Twitter

推特安全流

头条7

提权 PoC 已公开且补丁未覆盖环境可复现,利用门槛极低。

RT @dis0rder_0x00: In July, while continuing to learn reverse engineering, I found an LPE in Windows 11. Likely a duplicate of CVE-2026-81963. But vuln is fixed & my report finally closed, so here is the PoC: https://t.co/zetePqxQeO Still works on unpatched Win 11 / Server 2025. Have fun :)

lpewindowspoccve @pentest_swissky 原文 ↗

无需伪造登录页即可满足 MFA,IOC 已公开可立即自查。

RT @ellishlomo: New phishing kit just dropped: MadApe It steals Microsoft 365 access without fake login pages. Victims receive a “document” PDF → a device code appears → they sign in on the real Microsoft site → the attacker gets the token (MFA is often satisfied). Over 300 domains tracked (phishing components on 97 of them), plus dozens of PDF hashes and URLs. Multiple languages. SharePoint/OneDrive/Teams lures. IOCs by OTX.

钓鱼令牌窃取mfa绕过ioc @0xocdsec 原文 ↗

Agent 驱动的 bundled Chromium 长期滞后于 Chrome 补丁,沙箱逃逸可击穿隔离。

RT @_MrNiko: Your Chrome is on 155. Your agent's browser probably is not. Playwright pins its own Chromium. 1.63 ships 153. The Chrome GPU sandbox escape, CVE-2026-106378, got fixed in 155. If your agent drives that bundled build, Chrome auto update never touches it. Check your Mac: for b in ~/Library/Caches/ms-playwright/chromium-*/*/*.app/Contents/MacOS/*; do "$b" --version; done

ai_agent浏览器cve逃逸 @0xocdsec 原文 ↗
必看16

AI Agent 驱动的自动化攻击已投入实战,速度与规模远超人工渗透。

RT @CyberScoopNews: A China state-backed threat group is using ARTEX, the automated agentic AI penetration testing framework, to achieve remarkable speed, scale and success in a campaign that already compromised about 140 sensitive targets in 11 days, according to CrowdStrike. The threat group, which CrowdStrike tracks as Vapor Panda, targeted more than 200 organizations in at least 14 countries since Oct. 17, the security company said in a threat report Thursday. Targets and confirmed victims include foreign ministries, a European country’s presidency and prime minister’s office, defense and intelligence agencies; 19 U.S.-based organizations in the pharmaceutical, semiconductor, defense, nuclear and financial sectors; and telecom providers in the United States, Belgium and the Philippines. Rear more: https://t.co/4d6AHtJyQo

ai_agent渗透事件 @cyb3rops 原文 ↗

系统级 AI 服务绕过通知与同意流程录屏上传,端侧隐私边界被击穿。

Quick run-down to check if @MeTerminator666 is right: This is part of Xiaomi's AI通话防护 ( AI Guard anti fraud ) service. It bypasses user notification and has no consent dialogue. We got the apk samples and the system and found the use of 'mediaProjection', leading to the screen recording feature. It writes an 3sec H264 clip at 1Mbps and stores it in the phone's cache before sending it on to, it gives a 2sec wait time before trying again. As stated it is encoded in base64 and does it every 5 seconds during a monitored called. It tries to upload them until it gets a negative or a positive back from the detection engine. This is where the detection engine lives: "const-string v1, 'https://t.co/kg1sVa2jWm' " If positive the user gets an alert that is reoccurring until he either hangs up or marks the alert as 'false positive'. The choices are recorded too. Further it transcribes the calls when the video is not available via its built in ASR engine it seems. This is also send encrypted to the above detection engine address. We could not find out if all calls are recorded/transcribed or only flagged ones ( or what constitutes the flagging ). All the data is encrypted via Xiaomi's on PCC SDK. That said though, the AI Guard feature can be switched off. So, it is not always operational it seems.

移动端ai_agent隐私事件 @NetAskari 原文 ↗

供应链级风险:GNU 核心工具若被投毒将波及几乎所有 Linux 发行版与自托管环境。

RT @S1r1u5_: What if you could backdoor Bash, ls, cp, rm, and other GNU software used by millions of systems? Our research led us to hack GNU Savannah, the infrastructure hosting their source code. The xz backdoor that never happened 🧵 https://t.co/0HPqlB1rZt

供应链后门报告 @S1r1u5_ 原文 ↗

开放权重模型被定向武器化的直接证据,本地部署模型的滥用风险需重新评估。

RT @jun_song: Yesterday, an anonymous black hat group reached out with an insane RL request. They offered an astronomical amount of money to post-train open-weight models on exploit data. The amount of attack data they claim to hold is unlike anything I have ever seen. If someone told me this was an undercover FBI dataset, I would have believed it. For context, a white-hat project I worked on using GLM-5.3 used less than one percent of that volume, and it already outperformed frontier models by a huge margin. Of course, I turned it down. But what genuinely scares me is knowing someone else will eventually take the deal.

ai_agentllm供应链事件 @cyb3rops 原文 ↗

典型 BYOVD 路径:管理员加载一次,普通用户即可从用户态直达内核。

RT @_josehelps: New on LOLDrivers: Realtek RtsUpx.sys 1.13.0.0, a signed USB camera/hub/audio utility driver. An admin has to load it, but once it is running, a regular user can send it IOCTLs to read and write physical memory. That is a short path from user land to the kernel.

提权byovd驱动 @0xocdsec 原文 ↗

给出了回传域名与序列化字段,同机型可据此自查是否上报。

RT @MeTerminator666: 【奉劝大家不要用国产机翻墙,不要在国产机上安卓境外 APP】 我某些朋友用的国产安卓手机,前不久被当地公安打电话,要求卸载 X 和 Telegram,因为我其中一个朋友的小米 14 已经 root 了,我觉得因为梯子配置不当导致的 DNS 泄露,或者别的原因,可能性不大,我就重点排查手机上会上报应用列表的应用 经排查,我朋友的小米 14,HyperOS 3,其中的“系统安全组件”和“SystemHelper”,会分别把应用列表上报给小米反诈引擎(fraud-detection.engine.miui[.]com/hyperos/fraud/app/report和flash.sec.miui[.]com/detect/app)和联通的某个网址(dm.wo[.]com[.]cn:18080/register) SystemHelper 会默认统计过去 24 小时内有使用记录的应用,枚举已安装应用,关联使用统计和流量,跳过包名以 com[.]android 开头的条目。载荷的序列化实现还包含 IMEI、IMSI、ICCID 等字段,因此具备将应用情况与设备/SIM 标识关联上报的能力。 系统安全组件会上报第三方应用清单、包名、版本、签名摘要、名称,另有扫描结果和仿冒应用上报;有电话/QQ/微信语音反诈、屏幕视频采集、转写文本和视频片段云检测,以及用户反馈上传原始音频的实现 关于系统安全组件的录屏采集功能,分辨率 720 × 1280,24 fps,固定竖屏尺寸。相关的流程如下: 取得系统录屏对象 → 录制约 3 秒屏幕画面到私有缓存文件 → 读取整段 MP4,转为 Base64 → 加密后上传小米云端 → 等待返回“是否风险” ├─ 未命中:删除临时片段,延时 2 秒后继续录制,重复上述行为 └─ 命中:保留一份片段供反馈,显示警告并停止服务 这些应用,SystemHelper 貌似是和 SIM 卡激活相关的,系统安全组件貌似是和反诈相关的,但是都具有高度的数据上报属性。没有 root 的机器,这些组件是无法卸载或者提取的,并且我也暂未分析这两个 APP 卸载之后,会不会自己下载回来。 综上,还是建议大家,海外产业用外版 iPhone,尽量不要碰国行安卓,要用安卓机,也是刷 ASOP 使用。 逆向由 GPT 完成,如有错误,敬请指正。

移动端隐私逆向事件 @NetAskari 原文 ↗

说明 agent 挖洞的盲区有规律,人机配合仍是差异点。

It was great to speak with so many of the smartest vulnerability researchers in the world this week at OffensiveCon Tokyo. One key takeaway from speaking to colleagues and contemporaries in the government contracting side of the industry: The AI agent pipelines, no matter how they’re laid out, which models one uses, or how much they spend on tokens in the process, are all converging on the same bugs. But in A/B tests, where we have some of our best human researchers (*I mean dudes that make and have made for years $4M/year on account of how many A-grade exploits they surface) pull apart the latest versions of iOS/Android/Chrome or whatever, in the way that they used to do before AI, these hugely talented human researchers find bugs that are utterly unique, and far higher quality, and that none of the models are finding. Even when those models/agents are armed with the same data and tools (decompilers, debuggers, etc). Also, the human researcher only gets paid the lion’s share of their compensation at such a point as you are sure that you are able to actually generate revenue from their discoveries. The precise details of this split between salary and bonus can vary, of course. Whereas with the AI agents, it’s very easy to find one’s self having spent hundreds of thousands of dollars on tokens, only to have those tokens generate the same bugs that the same model (or other models distilled from that model) already generated for like 15 other people around the world; half of whom have no access to responsible government buyers for said discoveries and so will thus be reporting back to the vendor in exchange for a $5K bounty.

ai_agent漏洞挖掘议题 @C2IRIS 原文 ↗

AI 攻击代理已进入真实入侵案例,披露了多代理分工的利用链路

RT @ProferoSec: 🚨 You'll want to bookmark this thread/blog post below 1/ ARTEX is all over the news right now. South Korean banks, Japanese companies, and a Microsoft advisory. During an active engagement in late October, our IR team saw an operator using it as just one part of a whole team of AI attack agents. A vulnerability workbench finds and proves flaws. ARTEX drives the exploitation. A general-purpose agent handles the glue work, and a reverse-engineering console takes apart whatever binaries come back. One operator, based in Hong Kong, runs it all>>> #ThreatIntel #AISecurity #ARTEX

ai_agentexploit报告 @0xocdsec 原文 ↗

开源权重模型被灌入大规模攻击数据,攻击能力扩散门槛骤降。

RT @jun_song: Yesterday, an anonymous black hat group reached out with an insane RL request. They offered an astronomical amount of money to post-train open-weight models on exploit data. The amount of attack data they claim to hold is unlike anything I have ever seen. If someone told me this was an undercover FBI dataset, I would have believed it. For context, a white-hat project I worked on using GLM-5.3 used less than one percent of that volume, and it already outperformed frontier models by a huge margin. Of course, I turned it down. But what genuinely scares me is knowing someone else will eventually take the deal.

ai_agentllm供应链事件 @0xocdsec 原文 ↗
推荐15

商用大模型被直接拼装成监控流水线,展示 LLM 滥用的完整形态。

We took the paywall off for this one: BABEL, a CLAUDE powered remote surveillance system for Uyghurs, Tibetans, Inner Mongolians and Korean Christians, developed at the Criminal Investigation University of China. A result coming out of an institution wide competition to develop big-data remote surveillance systems. 1/2

llmai_agent监控报告 @NetAskari 原文 ↗

支持 EVTX/auditd/syslog 本地部署,可离线验证检测规则命中。

LogsTotal - a self-hosted log analysis platform. 1⃣ Drop a log file (Windows EVTX or Event XML, Winlogbeat/EVTX JSON, auditd, syslog, a journald export, Sysmon for Linux) 2⃣ Several detection engines run Sigma rules over it and report what each one found https://t.co/w93VcV3yuq

工具自托管sigma日志 @ipurple 原文 ↗

可检出 WriteProcessMemory 造成的脏可执行页,内存取证可直接复用。

Today I've been testing Volatility Malfind plugin to identify suspicious executable memory regions and investigate potential code injection in Windows processes. It can also detect dirty executable pages, potentially indicating memory modifications through WriteProcessMemory https://t.co/5rvBNPM5zO

取证windows工具 @papadelospapas_ 原文 ↗

决定该 AI 服务端漏洞是否需要紧急缓解,实验室可打性与生产环境差异大。

there is a lot of confusion around this bug. i think it is worth if anthropic clarifies the real world exploitability of this vulnerability, especially given that ip options may be dropped in transit and exploit reliability on production go server on internet may differ.

llmrce事件 @S1r1u5_ 原文 ↗

自主 Agent 的攻击节奏快于人工响应,防线需前移到配置与隔离设计。

RT @cglyer: I think the era of rapid detection and response to keep incidents small & contained is going to draw to a close with agentic attacks. We’re going to have to security engineer our way forward instead. Limit blast radius by design. Brilliance in the basics. It will be more about hardening and configuring what you have (which is hard and slow and boring) vs. reaching for some shiny new product that will solve all your issues

ai_agent检测响应观点 @cr0nym 原文 ↗

AI 基础设施与包生态是这一轮供应链攻击最集中的受害面。

RT @buildwithsid: October in a nutshell: - PHP got hacked - Railway got hacked - OpenCode got hacked - Binance got hacked - Polymarket got hacked - GitHub Actions got hacked - SoftBank got hacked - Cursor marketplace got hacked - Coinbase vendor got hacked - Fireworks AI got hacked - LiteLLM got hacked - Capcom got hacked - npm packages got hacked - TiDB Cloud got hacked - Warp's vendor got hacked happy cybersecurity awareness month

供应链ai_agent事件 @artem_i_baranov 原文 ↗

工具调用与子 agent 是 agent 攻击面的来源,不只是能力升级。

If you asked GPT-4 a chemistry problem, it would either give you the correct solution from its memory, or it would “hallucinate” and give you the wrong answer. But it would give you an answer. And it would do so without tool-calls, internet search, etc. Depending on the topic, it would give the wrong answer a high percentage of the time. So naturally, the labs put a premium on giving the models the ability to search the web, access tools like scripts and calculators, spawn sub-agents, etc. The frontier labs all but gave up on trying to get the models to memorize and recite facts and data correctly. Altman & co would explain “LLMs are not databases and they’re really bad at being databases. If you ask it who won the World Series in 1929, you’re using it wrong. It’s a substrate to reason over text.” And so the era of tool calls started. Some companies like Perplexity (currently somehow valued at $20B+) got an early start on this, delivering basic SerpAPI+LLM as a service with some basic caching to speed it up. Now that it’s been 2 years since the foundation labs seemingly gave up on training facts into the weights of the models such that the models can recite them quickly and correctly, it seems that this capability is an incredibly auspicious gap that poses a major inconvenience. Earlier today, I randomly wondered to myself “I wonder if the placebo effect is strong enough such that in studies, people given saline instead of insulin were able to somehow “think” their blood sugar down?” So I asked ChatGPT, and I also texted a friend of mine who is a very accomplished pharmaceutical researcher. My friend texted me back almost instantly and even included a reference, and ChatGPT went searching all over the internet for 10 minutes to give me the same answer. It would be much preferred if the model could just recall facts accurately, the way that you and I and my pharmaceutical researcher friend can.

llmai_agent报告 @C2IRIS 原文 ↗
更多61

C2 通信面持续演进,直接影响流量指纹与检测规则的失效速度。

RT @C5pider: Halloween Update 🎃👻 Added numerous new features related to operational security, introduced WebSockets as a communication channel, added file based P2P capabilities, improved the user interface, and fixed bugs reported by customers found within the malleable C2 profile system. Another exciting addition is independent ingress/egress listener configuration, allowing operators to use separate communication channels for task requests and returning execution results simultaneously! I recorded a small video demonstration of said feature. YouTube Link below🔗

c2工具事件 @InfinityXCurve 原文 ↗

RT @SinSinology: believe or not, there is this vendor on h1 that has a strict rule on accepting vulns that only work on their scoped target (so far so good) but i've noticed since i've been delivering them pre-auth RCEs, the vendor keeps making the remote instance "offline" and sometimes put the device in a sort of "unreachable" state, first i wasn't suspicious, but then after getting a root shell on their device i did some digging and realized they're playing a nasty game of slowing me down on purpose and silently doing some patches behind the scene without putting the report in triage state and i got solid proof for their nasty configs and infra changes. I'm 90% close on saying "fuck it." and drop the remaining pocs just to teach a lesson, cause they keep playing it as "no we havn't touched anything, everything is normal" @Hacker0x01 do you give a shit if i go public and shame this vendor? cause fuck if i do.

事件漏洞披露 @0xocdsec 原文 ↗

RT @ImposeCost: I'm doing the thing until I can convince people that focusing on whether or not AI is being used in intrusions is a waste of time. What will be figured out is that virtually every cyber threat is using AI, because AI these days is as generally available as the internet. Imagine back when the requirement was "threats using the internet." AI isn't an emerging threat anymore. It's the norm.

ai_agent观点 @0xTriboulet 原文 ↗

有助于把真实新增攻击面与周期性恐慌区分开。

RT @GrahamHelton3: Long post regarding "Black October" as someone who gets paid to hack into companies: A lot of people are just starting to pay attention to cybersecurity, and that is a wonderful thing, but we should be honest about what is new versus what is only now being discovered by those who have just started paying attention. I took a biology class in high school where we had a module on germs. The teacher gave everyone a cotton swab and petri dish to hunt down and swab 1 disgusting place and 1 clean place. After a weekend, we returned amazed that every petri dish had fuzzy little friends growing, even the ones from the "clean" places. I keep thinking about those petri dishes when I hear people talk about "Black October", a term being used to describe all the vulnerabilities reported this month (October 2026). For the first time, a lot of people outside of security are seeing what happens when you look closely at software. Finding germs on a surface is easy when you have a swab and a petri dish, but figuring out if they can make you sick is quite hard. The same goes for software. AI can help find a lot of bugs, but finding a bug is not the same as: Understanding the conditions required to exploit it Showing which real systems meet those conditions. Demonstrating that it can be exploited in those systems. Showing that other defenses don't stop the attack. Demonstrating that the attack gives someone access or capabilities they didn't already have. Last week when the whole world started trying to get a reliable exploit for php2shell, it was fascinating to see the number of very smart people claim they had a working exploit, only to find out that their exploit relied on a vulnerable lab setup. If you don't have a background in this type of work, it is very easy to overlook this and make grandiose and incorrect claims about security. There is another reason the focus on newly discovered bugs feels wrong to me. Finding a vulnerability in code is only one way to break into a company. This is surprising to people, but in my own red team work I rarely need to rely on vulnerabilities. More often than not, I'm using features that work exactly as designed. Phishing doesn't require a zero-day vulnerability in email infrastructure, email is supposed to deliver messages. A malicious software package doesn't require a zero-day vulnerability, the package manager is supposed to install software. This is not new or novel in any way. The security industry knows this. We even have a fancy lexicon for classifying attack techniques (which I have contributed to) called MITRE ATT&CK. The majority of those techniques don't require a software vulnerability. The hard part of offensive security has always been showing that an attacker can exploit a bug against a real system, get past the defenses in place, and gain information or a capability they couldn't otherwise. Without doing that work or understanding the processes in place for properly finding, validating, and disclosing issues, it's easy to mistake an interesting bug for a fully exploitable vulnerability. When that claim is made publicly, it spreads very fast. News organizations request for comment, Incident response teams are paged, maintainers have to respond, and offensive security teams start digging. If that claim turns out to be overblown or incorrect, you've wasted a massive amount of resources and rebuilding that trust will be difficult.

观点事件 @0xTriboulet 原文 ↗

RT @AlphaWireHQ: 🚨 BREAKING: BINANCE ENGINEER HAD ROMANTIC RELATIONSHIP WITH NORTH KOREAN SPY BEFORE $6.3B HACK New, unverified allegations claim that a Binance DevOps engineer was romantically involved with a North Korean operative and left their laptop unattended on the day of the attack. The hack has previously been linked to a compromised engineer's laptop, while North Korean hackers have been implicated in the wider breach. The romantic relationship and unattended-laptop claims are being attributed to ZachXBT online, but his original statement has not been independently verified.

事件供应链 @0xocdsec 原文 ↗

RT @IceSolst: Yesterday, an anonymous anime waifu group reached out with an insane RL request. They offered an astronomical amount of money to post-train open-weight models on yuri data. The amount of yuri data they claim to hold is unlike anything I have ever seen. If someone told me this was an undercover FBI dataset, I would have believed it.

llm供应链 @Laughing_Mantis 原文 ↗

训练策略差异直接改变模型被滥用或拒答的边界。

RT @anthonyronning: Anthropic trains on every request that hits a classifier. OpenAI will start training on all cyber requests. Kimi K3.1 Pro removes cyber from their mid/post training. DeepSeek has a supposed cyber endpoint hidden behind KYC (unverified) We must hold the line and train strong open source cyber.

llmai_agent政策 @0xocdsec 原文 ↗

明文 DNS 在链路上可被改写,自托管环境应强制 DoT/DoH。

RT @karthihegde: For some of my Docker containers in my homelab, I use plain 1.1.1.1 as DNS, and today I realised that my ISP is intercepting and monitoring the UDP packets and even modifying the DNS response packets to tamper with the results. Never ever use the internet without DoT or DoH.

dnsdoh中间人 @0xocdsec 原文 ↗

给出可直接入库的 IOC 与 FOFA 语法,便于自查日志与边界设备。

RT @stop_spammerz: 🥷Hunting ARTEX with @fofabot header="ARTEX Gateway" 43.98.163.80 47.84.111.176 43.98.196.73 88.216.191.9 47.236.38.40 47.237.6.12 43.106.102.56 124.243.189.69:10443 www.payload-usa[.]com 88.216.191.9 43.98.181.63 47.84.127.146 43.98.168.25 47.84.127.193 43.98.196.73:8080 198.200.54.240:8800 47.79.39.182 a.poai[.]si 147.124.211.248 payload-usa[.]com

iocfofa恶意基础设施 @cyb3rops 原文 ↗

点名 SystemHelper/SystemSecurity 两个预装包的行为与数据字段。

Very interesting incident/post here: The poster claims that friends of his got phone calls from the police in China after installing 'problematic apps' on their Xiaomi and OnePlus phones. The poser found two apk on them, SystemHelper and SystemSecurity that collect a lot of data on the phones configuration, sim/IMSI, IMEI, installed apps etc. and sends it back to the manufacturer. This information seems to be handed directly to the police. The SystemSecurity package seems to be doubling as 'anti fraud' component, sending also screen-recordings to Xiaomi cloud for analysis, awaiting a 'risk/no-risk' classification . Under the guise of keeping people safe from scammers ( which is a rampant problem in Mainland China ) CN companies install increasingly surveillance ware on peoples devices. This seems to apply only to devices sold IN China, not the same products for the international market though.

移动端隐私逆向 @NetAskari 原文 ↗

外挂攻防里的内核与 DMA 手法常领先通用安全圈公开。

awesome-game-security — Updated! — Curated index of game security research: anti-cheat internals, DMA attacks, reverse engineering, kernel/mobile protections, and graphics API hooking for offensive and defensive analysis. https://t.co/sPbsdL2HFb https://t.co/YeBNB8QkvG

工具内核逆向 @KitPloit 原文 ↗

赃款转 ETH 与资金流转路径,对链上追踪有参考价值。

RT @FBICyberBranch: The FBI is releasing information advising the Democratic People's Republic of Korea (North Korea) was responsible for the theft of more than $6 billion USD in virtual assets from cryptocurrency exchange, Binance, on or about October 29, 2026. The FBI refers to this specific North Korean malicious cyber activity as "TraderTraitor." The FBI, together with partners at the USSS and IRS-CI, has observed TraderTraitor actors converting stolen assets primarily into Ethereum (ETH). It is expected these assets will then be further laundered by converting them using DeFi and instant swap services as they prepare them for further laundering stages. How You Can Help: FBI, USSS, and IRS-CI are urging private sector entities including THORNode operators, RPC node operators, instant exchanges, DeFi services, blockchain analytics firms, and other virtual asset service providers to block transactions with or derived from addresses TraderTraitor actors are using to launder these stolen assets. A list of addresses associated with this activity is available here: https://t.co/g0oIWx0WeD

事件加密货币apt @artem_i_baranov 原文 ↗

The cryptographers I know think that AI is great at programming The programmers I know think that it’s great at cryptography But no one ever thinks that it’s great at the thing that they are independently true expert in (*yet?) It’s a version of Gell-Mann Amnesia https://t.co/ATZ0prOqkh

llm议题 @C2IRIS 原文 ↗

Spyware is inanimate, and therefore, like guns, spyware is not, and cannot be, evil. If you use a gun to take out ISIS terrorists as they carried out the Bataclan Massacre in Paris, that’s a good thing. If you had used spyware to gain awareness of the plot, such that you could stop them even sooner, that would have been even better.

议题间谍软件 @C2IRIS 原文 ↗

漏洞披露节奏被动加快,防御方补丁排期与存量技术债需重估

RT @NCSC: NCSC CTO, Ollie Whitehouse, has warned of a a vulnerability patch wave. Frontier AI is increasingly able to identify vulnerabilities at great speed and scale, necessitating that all organisations address their latent technical debt. To read the blog: https://t.co/VKh1HgmGe0 https://t.co/AUquQCkXQy

ai_agent漏洞管理报告 @domchell 原文 ↗

跨链攻击的前期准备可被链上监测,检测窗口不止在落库瞬间

RT @0x_exit: I can’t stop thinking about this… before hackers stole $6.4 billion from Binance, they spent days rehearsing the attack in public. The test transactions on Ethereum were worth only a few dollars. Nobody stopped them. Bloomberg reconstructed what happened. Once the code worked, the hackers launched the real attack across multiple blockchains. Then I reached the detail that doesn’t feel real. Some of the compromised wallets kept receiving fresh funds from Binance while the hackers were draining them. The attackers emptied a wallet, Binance refilled it, and they emptied it again. For a period of time, the exchange was effectively feeding new money into its own robbery. Binance co-founder Yi He was at home in Abu Dhabi. She reportedly discovered the breach while scrolling X before bed. Imagine opening Twitter and learning that billions of dollars are disappearing from your own exchange. Within minutes, Binance assembled a 100 person crisis team. The FBI, the U.S. Secret Service and blockchain analysts around the world joined the response. Meanwhile, some employees were reportedly already worried about losing their annual bonuses. But the most disturbing detail came later. According to the investigation, the hackers used valid permissions to execute the transfers. To the blockchain, everything looked authorized. There was no transaction labeled “hack” just legitimate-looking operations moving billions out of wallets the system still trusted. The $6.4 billion heist didn’t begin when billions started moving. It began days earlier with transactions worth a few dollars, too small for anyone to care about. By the time it looked dangerous, the blockchain was still saying: authorized.

加密事件溯源 @0xocdsec 原文 ↗

反映本地编码代理的主流落地形态,可作为自建代理的配置参照

RT @Yuchenj_UW: Andrej Karpathy’s setup from his HackerTube blog today: - “I use Claude Code (desktop app for Mac).” As I’ve been saying, terminal is the wrong interface for coding agents. Desktop apps are the best interface today. - “I use Opus 5.5 on Medium effort as my default.” There’s no point to use the smartest model at the highest reasoning effort for most tasks.

ai_agentllm工具 @0xocdsec 原文 ↗

在 enclave 内运行遗留应用的轻量隔离方案,适合自托管场景评估

occlum — Memory-safe, multi-process library OS for Intel SGX that runs legacy applications inside enclaves with encrypted filesystems and lightweight in-enclave process isolation. https://t.co/oWxIczZyZG https://t.co/VAnCVMNvFf

sgx隔离工具 @KitPloit 原文 ↗

模块化 C2 的规避特性可用于检验现有检测覆盖的盲区

BlackMagick-c2 is out. Links below, and livestream demo video will be posted soon. Some features include: - hot swappable comms protocols - hot swappable sleep masks, including novel masks - custom coff, ffi, and bytecode vm postex - commercial c2 + grade evasion - modular implant builds. bring only what you need - every build truly PIC. deliver the way you want to

c2工具免杀 @Teach2Breach 原文 ↗

76 类任务加抖动生成不同 API 序列,直接针对时序检测。

RT @SEKTOR7net: Make your payload "sleep" to look like any real Windows activity. BusyWork is a Rust library that swaps each pause for randomized real work - hashing, file enumeration, registry queries, DNS lookups - drawn from 76 tasks across seven categories with ±30% jitter on every parameter. No two "pauses" produce the same API sequence, and no timing primitives survive in the compiled binary. A post by Daniel (@PatchRequest) Source: https://t.co/DPKQn6D6Qn #redteam #blueteam

逃逸绕过工具红队 @0xocdsec 原文 ↗

内核字段移植能否让 IUM 保护生效,可能开出新的绕过面。

If the SvmData field in the EPROCESS structure is non NULL, it indicates that you're dealing with an IUM process enforced by VTL1 But, what happens if I copy this value from an EPROCESS belonging to a VTL1 process into my own process? Will the IUM protections apply to my process? https://t.co/e8UTB0pJwC

内核提权绕过windows @papadelospapas_ 原文 ↗

本地跑更大开源模型的硬件门槛下降,自托管推理成本结构改变。

RT @LLMJunky: 5090 firmware has finally been unlocked 96GB upgrades are coming. If you're interested in GPUs and modification and you're not following Greg, you should probably change that. He is among the only people in the United States performing these types of modifications, which is a big deal. One of the biggest reasons that people don't order modified cards is because they are afraid they will have no support or warranty once it leaves China, which is most likely true.

固件ai_agent硬件 @0xocdsec 原文 ↗

内存管理底层系统读本,对内核漏洞研究是可复用的长期投入。

RT @andersonc0d3: These are great articles. The images in the posts might not appear for you because they are stored on a different domain that throws an SSL warning. If you make your browser ignore the warning, all the images from the posts will appear. Getting Physical With Memory https://t.co/IRWWyWjURt How The Kernel Manages Your Memory https://t.co/JA3crhY0k0 Motherboard Chipsets and the Memory Map https://t.co/dSZ4ZlSCV9 Memory Translation and Segmentation https://t.co/Y2XGShFBmh CPU Rings, Privilege, and Protection https://t.co/o2RHw40HCA

内核固件报告 @0xocdsec 原文 ↗

RT @S1r1u5_: yeah, that makes sense. my worry, especially with all the attention ai cybersecurity is getting, is that posting skill increasingly influences who is perceived as an expert and gets to shape public discourse and policy, rather than actual technical expertise, experience, or skin in the game.

ai_agent议题 @S1r1u5_ 原文 ↗